There's two basic kinds of distillation: 1) the massive [and dumb] method where you ask a question and use the answer as reinforcement (Black Box), and 2) more targeted distillation where you use one model to directly inform/train/guide another model (RLAIF).
The latter is basically fine-tuning the model with direction from another model. Thousands of businesses do this every day to fine-tune. This is almost certainly what the Chinese labs are doing, since it has a much better effect on the end result than just getting simple answers to simple questions.
These complaints of distillation are inflating the problem to make it sound worse than it is, because they want the USG to block/ban Chinese model providers as protectionism. They have already called for more export controls on chips (which is funny because DeepSeek v4 was designed to run on Huawei chips and now the other Chinese providers are following suit). But they can't come right out and say that, so their claim is that they're asking for more export controls because distilled models might not be as safe as their own. But if you show them a jailbreak of their model that bypasses their safety, they'll tell you that any model can eventually be jailbroken so don't worry about safety.
> These complaints of distillation are inflating the problem to make it sound worse than it is
Unfortunately, the Reuters piece itself is complicit in this dramatization. The lede paragraph parrots Anthropic's talking point that distillation is an "attack", without using quotes that would alert the reader that this framing is a corporate talking point. Distillation is NOT an attack.
ironically, I think this is why the jobs apocolypse is overblown, Ai is only good at a thing if the people using it are also good at that thing, and people are attributing Ai as superhuman at things they do not know themselves
Has it been proved in a court of law that it is a copyright violation?
In some cases if the model regurgitates the original material then that is clearly copyright violation, but if the model "learns" from the source material just like a human brain would then that's not a copyright violation.
No, what was proved in court was that they downloaded and trained on millions of pirated books. The court said their use of books is fair use, but stealing them isn't.
I think we're going to see cases that find distillation is also fair use. You're using the competing model like a book. You pay for it, you use it (read it), it informs your model, but you aren't repeating/reselling what the model told you verbatim. Foreign labs may still run afoul of competing labs' Terms of Service, and they may also pay a settlement (or not, it's a different jurisdiction after all), but the damage is already done. Distillation will become uncontroversial when done legally.
> Has it been proved in a court of law that it is a copyright violation?
God I'm so tired of this.
The billion dollar companies have the ability to hire an army of lawyers to DDOS the legal system. They at most pay a slap-on-the-wrist fine as the cost of doing business.
I'm extremely pro free markets etc, but the uncomfortable truth is anthropic stole the work thousands of authors for profit. I think it will one my favourite things in life: programming books.
it's a 'too big to fail' model. Because they have a big swinging dick all the copyright and other restrictions they violated would nuke them from orbit so we can't actually hold them to account for it .... for some fucking reason.
even if you disregard training costs, pure inference costs are a problem same reason other api have rate limit. this is an attack to bypass the rate limit.
Be careful to properly identify the bad behavior. A customer who buys a product for less money than it cost to produce has not necessarily done anything wrong. They just took advantage of a loss leader. That's on the seller.
Did you notice that when Valve was displeased about scalpers, Valve changed Valve's behavior?
It doesn't seem reasonable to complain that a customer of your AI service received that service for less money than it cost you to provide that service. I don't think that is the complaint here at all. If that was the issue, they could just raise their price.
As most everybody seems to notice, this is just a reenactment of what was once written for comedic effect: "You're trying to kidnap what I have rightfully stolen!"
Still calling it an "attack" feels like a stretch.
They literally had to pay for that "attack", no matter how many accounts they used.
Google was killing many websites for decades with their crawlers. Most large websites decided to create dedicated infrastructure for their traffic alone. Somehow they didn't participate in that cost and were not called the attackers.
This is the mental mental leaps I'm struggling with here. Did you not live through that era where they were explicitly and repeatedly called out as 'attacks'? They were generally tolerated/hardenee around as they provided value-in-discoverability.
Just to ensure you don't gaslight yourself - I did live through that era and I worked on and supported a niche community (a MUD) where we did a lot of work encouraging marketing and discoverability through MUD forums as well as making sure our page was accurately and minimally keyword tagged and highly available for indexers.
In the time since that era search engines have transformed into platforms themselves that do engage in more parasitic behavior but it's important not to assume that the way it is now is how it always was - that's a rather defeatist path to walk down where you ignore awareness of the fact that there can be a highly profitable non-enshittified search engine that supports, rather than destroys, the ecosystem it benefits from.
It was better and, if we're diligent, it can be better again.
They should be. But as the saying goes, one website/company dying is a "tragedy," lots of them dying at the hands of one company is a statistic of corporate growth. Or something like that.
And then of course when the tables turn on a company and they're the ones getting bombarded, they cry foul. Keep in mind Anthropic did many similar things that you mentioned Google did.
I think the term "attack" here is appropriate but not in the way Anthropic is framing it. Alibaba is clearly violating terms to extract data, so that's definitely not above board. But it's not like a DDOS attack where Alibaba is trying to attack Anthropics servers. Alibaba is simply doing exactly what Anthropic did to the rest of the internet, just targeting Anthropic and paying them to do so.
My terms of service are that you are not allowed to breath oxygen.
I am getting a bit tired of companies being able to have user hostile, anticompetitive, monopolistic terms of service. The freedom we give them comes at the cost of the freedom as consumers to have free markets because they lock them up
Illicit means maybe against the law but definitely against the rules, for example an illicit affair. The word for against the law is illegal, from Latin, or unlawful, from Germanic. I guess the Germanic cousin of "illicit" would be "forbidden."
Extramarital affairs are against the law in many countries and 17 US states. “Illicit affair” is potentially a holdover from when it was illegal more places, not just a conflating of against the rules with illegality.
That's violating TOS, spamming, possibly a DDOS, but the distillation in and of itself is not an attack it's just using the model.
Like the difference between scraping a site with one or two active connections vs thousands. It's not the scraping that is an attack, it is how they are going about it
Just sending a request to a service does not constitute an "attack". It seems that what Anthropic mean by "fraudulent account" is probably just one violating their terms of service - misuse of a subscription account, and/or the presumed nature of what the user was trying to do.
I guess Anthropoic would regard any developer using their subscription plan with OpenCode to be operating a "fraudulent account", maybe an "attacker" too. Now we know how they think of anyone using Claude to develop software competing with Anthropic. Only an "attacker" would want to vibe code their own harness, or god forbid want to learn how to build/train an LLM.
Of course Anthropic's wording is intended to be deliberately provocative, since they are trying to manipulate the US government into shutting down the Chinese competition.
Is an attempt to copy all or parts of a model an attack, when models have very questionable copyright status? Maybe? I don't think most people have much sympathy here though.
Let’s not forget that by the same logic, Anthropic et al are “attacking” copyright holders all around the world by scraping their data unauthorized for training.
Distillation done via bulk automated activity of fraudulent accounts, in violation of a terms-of-service, can reasonably be called a "an attack" – specifically a "distillation attack" – even though distillation itself isn't necessarily an "attack".
This is similar to how compromising an account through bulk automated trials of many passwords is reasonably called "an attack" – specifically a "dictionary attack" – even though using a dictionary is not itself an "attack".
You shouldn't need to smuggle your sympathies (for the tactic or perpetrators) or antipathies (for the target) into peculiar judgy language prescriptivism against common, understood usages.… that then label Reuters "complicit" for simply reporting Anthropic's claims accurately. That's what Reuters is supposed to do, in a story about a letter Anthropic wrote!
Labeling it as an attack is smuggling sympathies. It is not common; there are only a small number of people who even discuss the concept. A company buying a product with the intent to reverse engineer or copy its features is likewise not an attack; it's just normal competition that benefits society.
The standard of neutrality that people here pretend to require from news organizations is not even remotely realistic.
It was a timely story from Reuters. They do fast news feeds, like APnews. Could it have been better or more accurate? Sure, they could have gone into why distillation may or may not be seen as "an attack". But then it would have been a more involved story, defeating the purpose of a news feed.
The Reuters piece was "good enough". Some other place like the NYTimes or WSJ can follow up with more detailed investigative coverage if it's a worthwhile story.
I don’t want or need fast and “good enough” news and i’m gonna try and make a case that you don’t either.
Until very recently, all of modern civilization was built by people who got their news at most once a day. Reputable bureaus like Reuters took that day to get it right.
I’m not the national security advisor, so I don’t need a push notification that there was an earthquake in Nepal, or a bullshit rush-job briefing on Chinese AI distillation tactics.
The fast part isn’t for your benefit, primarily, and news media would love to go slower and have more time if they could, and still survive. The race to break news first - in order to be the one to tell their audience something “new”, something they hadn’t heard elsewhere - is real and it has been around for all of modern civilization, for hundreds if not thousands of years. A one day turnaround was a thing purely due to daily newspaper print runs being the fastest distribution, it wasn’t because it was long enough to get it right. The reason they had a day is because the competition couldn’t get something out faster than that. Then for a while there were twice daily print runs to be more competitive. Then the internet came along, and now the only way for a site to get attention and be talked about on Hacker News is to report it before any other sites do.
There are some news media that do go slower and take their time, but I think they’re struggling to stay alive. Reuters is still reputable, but they no longer necessarily take a day. The big question is how do we get humanity to prefer slow & correct over fast, and it is even possible? When you hear about an earthquake in Venezuela, how do we stop people from Googling it immediately, and get them to wait for the best most correct story rather than reading whatever’s available now? In the case of natural disasters, I don’t think it’s possible anymore, no matter what case you make. I’m not sure it’s possible with stories like AI distillation either, even if you can absolutely cement the case for slow news. The fact that it’s async/internet now and that first still counts means we (you and I) are still going to give traffic and attention to sites that have the first information on a breaking topic, statistically, despite having a preference for correctness over speed. The one thing we can do is vote with our dollars by subscribing to whatever news media that does a better job than others.
Yes. It was good enough to communicate that news item.
Did Alibaba perform "an attack" or were they taking advantage of resources and going beyond Anthropic's terms of service? Didn't Anthropic do the same kinds of things when building their models?
These are all interesting questions, but they don't have to be addressed in full by a news blurb about a letter Anthropic wrote to some senators.
Distillation may not be an attack, but it is a ToS violation and could be seen as IP theft.
Any reasonable company would be pissed if a competitor, especially at Ali Baba's size, leveraged that company's R&D to compete. It is in this sense, a corporate attack.
If you want to roll your eyes at distillation concerns, you might need to excuse Anthropic for originally using pirated material to train their models.
More the opposite - companies who stole IP for their own benefit have no leg to stand on when others do it back. Personally I couldnt care less if Chinese labs rip off Anthropic. Its what America would do if they wanted to, for whatever reason (they probably do it right back secretly anyway).
> how would you feel if somebody quoting you would turn your word dramatization into "dramatization" because they don't agree with your assesment
This is exactly what news agency should be doing though. When the dude showed up to Comet Pizza to look for Hillary Clinton or whatever, do you figure they should've printed "Local hero saves children from predatory cabal"?
I want them to report the facts, not their opinions.
Reporting that corporate called it attacks is good. I do prefer direct quotes.
However, when they quote one word, the journalists are inserting their own opinion about it. I want to make my own opinions based on the facts. I don't need the reporter to draw the conclusions for me.
Well, let’s say you put the picture of some political figure, and put in highly contrasted red, bold large catchy font, "TERRORIST THAT KILLED MILLION PEOPLE", then below that in barely visible contrast, in tiny discrete letters, "is what this person probably will claim to be against".
This whole sentence technically will be correct, 100% guarantee, whatever this person actually even said or think.
From a propaganda point of view, framing the elements of language is even more important than what the statements actually states to be true or possibly true.
nice slippery slope you manufactured there - what if Reuters becomes Daily Mail
what framing are you talking about? they are literally quoting a company.
please explain what Reuters should have done here. Should they have added in parentheses: (editor note: we don't agree with Anthropic calling this an "attack")
Is that what you want? News outlets giving their opinion and moral judgement on company quotes? I mean, Fox News/CNN do have a large following, so there is clearly a market for that.
> please explain what Reuters should have done here
This is very straightforward: use direct quotes or use neutral language. The article describes the alleged incident as both an “attack” and a “strike” in the first two paragraphs. And neither is within verbatim quoted text.
Reuters, however highly you may regard them, simply adopted Anthropic’s framing uncritically in this instance.
If you’re going to call out their use of slippery slope as a fallacy then it should be pointed out that your original argument was framed on an appeal to authority of Reuters as a leading news agency.
I'm more excited by open weights models you can't self host and need to spin up on H200s (RunPod or bare metal). This is where the real power lies and is where the open source world will trend.
It's far cheaper to spin up an H200 hourly or to simply consume a managed version of an open weights model than it is to use a proprietary hyperscaler API. And you own the model itself and can fine tune, tweak, lobotomize, etc.
The stuff you can run on your own RTX cards is neat, but it's rather hobbyist. The real power is in the cloud. Renting cloud hardware is fine, because the core problem is ownership of the weights, not the server rack or ISP fiber lines. Those are already commodity.
Big businesses will eventually run open weights models in the cloud, and it'll be a rather large part of the future AI economy.
Eaaaaasy now, the Chinese labs aren't freedom fighters on behalf the common man. They're not non-profits, they're not neutral transnational organizations only dedicated to open source efforts.
They're Chinese companies offering open source models now as loss leaders to keep themselves in the game because they know virtually nobody, especially in the corporate world, would contract with them and give them access to their data. They might as well just send a Dropbox link of all their sensitive data directly to their Chinese competitors, same end effect.
They're also doing it as the digital equivalent of what they've done in other industrial sectors for decades. Undercut and flood the market and once you've killed or severely hindered your competition, then you have the market cornered. The moment they can afford to these open source releases will stop.
Then the world will be stuck, just the way the world is largely stuck on rare earths. Instead of being able to regulate the leading companies from DC and Brussels, they'll be stuck watching Beijing call the shots.
That world would likely always have guys like Mistral and Trinity, but it's an open question if they'll ever catch up to the frontier.
And then Beijing will enjoy access to the data (ask any multinational operating in China for more than 2 seconds how useful contracts and Chinas legal system is for protecting IP), and these companies will roll in the money, and the Chinese supply chain will grow up behind the labs.
So, let's not pretend they've got the moral high ground. No. That boot just isn't on your neck yet. They're playing the long game -- and they're good at it.
I think most of us know why they're doing it.
We are just very pleased with it regardless.
1. I get great products for nearly free
2. Anthropic/openai/etc will hopefully be destroyed since they stole everyone's work and are trying to capitalize on pure theft.
Win-win. The why of it is not really that relevant.
If you don't think Anthropic and OpenAI are multi-trillion dollar militarized behemoths you need to catch up on some news.
Both are planning $trillion+ IPOs this year. OpenAI is collaborating with the Department of War, and Anthropic is under intense pressure to do the same and their top model is being held hostage right now. This week, the Department of War wrote a statement that xAI should not be held accountable for environmental laws because Grok is a vital weapon system of the US and was used to fire over 2000 missiles at Iran. The pentagon's statement mentions there are 3-4 such models so you may be able to guess which they are.
> You don't trust the multi-billion dollar behemoth, but you trust the militarized multi-trillion dollar behemoth to play 'robin hood'?
Nobody's trusting anyone, we're just enjoying the benefits of true competition much like the working middle class gained benefits between the ideological competition of the Cold War.
The Chinese companies don't have to be open weights, and it's not all about competing with the west. For example, most of Ziphu's (GLM) business in China is supporting private on-prem instances rather than selling API access. They make money by selling support services - much like RedHat's busines model.
It doesn't matter why Chinese firms are stealing models and open sourcing them. The fact that they are doing it is a very, very good thing for basically everyone other than the people who paid to build the original models, but I've got no sympathy for them considering they stole all the content to train them in the first place. This is some kind of beautiful irony.
> it is a very, very good thing for basically everyone other than the people who paid to build the original models
It's not a good thing if you think there's more discovery and progress to be made, rather than cannibalising a fully mature field with cheaper alternatives. Drowning R&D early is not good for everyone.
What does further progress get us? Mass unemployment? Extinction? Pick your dark future science fiction?
The happy ending where we're all living in a garden of eden cared for by benevolent AI is hardly worth considering when you look at the cast of characters who are in charge of the world right now.
Is leveraging an enormous capital advantage to strip-mine the Internet and sell it back to us cannibalism or not? Confused on this point. I think they are exploiting a loophole in copyright law (and kind of redefining the meaning of "derivative work" in my opinion, but hey I'm not a lawyer) that collectively we tolerate because the end result is so useful
I think that's a slightly different topic, but: a) strip-mining the internet is definitely the most misleading way to think about it. Strip mining means aggressively removing something to the area's detriment, and nothing has been removed. If all AI is turned off today the internet has not lost all of its natural resources, and silly phrases like that fuel inappropriate emotions and consequent conclusions and b) the internet is not being sold back to us - that is also a highly misleading phrase, if not an outright lie. The internet is still there and we can use it. No one is selling back to us what we already had. AI is not the internet cordoned off and resold.
Can you please tell my, as someone who is neither Chinese nor American, "why" I should care if a Chinese company stole from another American company (that in turn stole from everyone) to give me a cheaper service that fits my use case?
> to give me a cheaper service that fits my use case?
Because they aren't giving you a cheaper service that fits your use case.
Best Case scenario, it's a trillion-dollar behemoth stealing from a billion-dollar behemoth so they can add their own explicit restrictions/weights on top to influence the masses.
There is no 'robin hood' here, any perceived value you get is clearly and explicitly tainted. "I don't care if it doesn't show me non-party-line results - It makes me a cheap UI !". Ethics/morals be damned.
> There is no 'robin hood' here, any perceived value you get is clearly and explicitly tainted. "I don't care if it doesn't show me non-party-line results - It makes me a cheap UI !". Ethics/morals be damned.
I can't tell if you are talking about Anthropic or Alibaba here.
In a world which already has the likes of Anthropic and OpenAI, having Chinese labs be a counter balance is decidedly better than the hypothetical where American companies had a global monopoly on LLMs.
If your argument is that all present LLM offerings are unethical then that is something I am sypmathetic to. That said, I am also unable to offer a conceivable roadmap to undoing the opening of the LLM Pandora's box so I tend not ground my arguments in anti-LLM advocacy; that would be very 2023 of me.
The main problem is how they accessed the IP, but then using it to train a model is fair use. But yeah, IP theft doesn't exist because nothing is stolen really: Hollywood studios still have their movies.
Um, yeah. They stole the IP and then they stored the pirated IP. It was literally stolen and stored on their servers. That proves that IP theft exists. It's not complicated.
I don't think that's true. Sometimes the 'why' is lost in time as no one's around to tell it, so we end up with a "if a tree falls in the woods and no one's around to hear it, does it make a sound?" scenario. It doesn't really matter. The thing now exists without a 'why.'
They want to create a monopoly and destroy every competitor, before they got a chance to rival them.
Why can't OSS software rival closed source software? It should be an open market, at least "somewhat", what's happening for real? EU providers will also get banned, if they reach or exceed US model capabilties?
Closed source providers can close your account at a whim like and destroy your business and then use the data you supplied them to create a competitor (Meta, Google, OpenAI, Anthrophic).
Well Zai's GLM 5.2 legitimately is a frontier-level model, though not quite parallel with Opus or Fable. Unfortunately, its too damn big to run locally for most people. Thats the bottleneck right now; the open-weight models exist but something capable of competing with the frontier models just can't run on anything normal yet.
https://research.nvidia.com/labs/lpr/slm-agents/ - Distillation data is a natural byproduct of using these models. There's no effective defence against it. Anthropic is degrading thinking blocks to summaries to slow it down and hide model internals, but in the end, the math says you're SOL and it works on MNC/Large Corporate scale well enough that the moment cost becomes a priority, you're left without the lock in you need to keep customers paying.
Byproduct? It’s essentially the only part of an LLM that is useful, because it’s the WHOLE product!
It’s the same reason why DRM for audio and video is a non sequitur - if you want a person to see or hear audio or video, eventually at the end of the chain, it’s going to be converted to audio for the ear and light for the eyes - that’s why you attach your tap.
Without a model generating tokens, what’s the point. So if Anthropic somehow disable quality token generation, what’s the point!
That's why the harness is moving server-side: because generating tokens is not the actual point of the model, not for the users. Especially with tool calling giving us agents that can act, most of the tokens generated are not, themselves, critical to the end users. Specifically, a lot of tokens goes into orchestrating actual tool calls, and then most "thinking tokens" are only relevant to users only in so far as they help users keep track of and verify what the LLM is doing. So all those tokens can be hidden or replaced by partial summaries, and all of that can happen server-side, and then there's very little to distill from.
Heck, one of my favorite fine tuned copies of Qwen uses Opus 4.6 Reasoning distilled. I'm not sure where the maintainer is based out of, but me in the states, if I had the hardware to do similar things I would. Its like you say, basically everyone is doing it. It kind of makes sense to me too given that you can have roughly similar data, but your reasoning logic is what the real secret sauce is in my eyes. It doesn't matter if you know everything in the world, if you don't know how to reason with that information.
It's about training data and using Claude to compare 2 outputs and have it indicate the better one. This gives you higher quality training data that you can use to train a fresh set of weights. Weights don't get adjusted on-the-fly, instead the dataset for training is improved and then you train a'fresh. And it's hard to detect because you're just asking the model which of these outputs for a given prompt is better? Or something along those lines.
Stupid question: I was under the impression that these models were trained on PB of data. Surely the amount of questions/response they can extract from querying a bigger model (Claude) is fairly modest. How is it not a drop vs the training dataset?
It's not about how big your dataset is - it's about how you use it.
I jest, but I'm also completely serious. 1T tokens from Claude can teach a model something 1T tokens scraped from the open web can't. Things like "how an LLM can problem solve effectively", or "how an LLM should use tools", or "how to construct reasoning chains", or "when to double check", or "what innate capabilities an LLM can or can't rely on".
Those are valuable things that Anthropic's own team spent a lot of effort post-training into Claude. Distillation allows them to be extracted and transferred to an otherwise unremarkable base model.
Base models have a lot of capabilities - arranged in all the wrong ways for high performance reasoning and problem-solving. The power of fine tuning on "a couple thousand of input-output pairings" is that it can fix some of that. If your pairings are very well chosen, that is.
Why? They often don't make sense. They send DMCA takedowns over materials they can't even copyright, for example. They fessed up to creating shadow libraries that they didn't even use in their training corpus, resulting in the largest copyright settlement ever. Your reasoning is flawed.
Most research converges to the idea that RL on synthetic data makes models worse, not better.
If what you claim was anywhere near that relevant, than we would've long achieved singularity by simply feeding increasingly better output to the training of the next model in a loop. Yet this doesn't work.
25 million turns on Claude output is a small amount, yet an expensive one (we talking hundreds of $ millions) that is better spent on compute.
There's no evidence such a process works, but I'd like to know more if I'm wrong.
Back up what? That distilling from a more capable model into a less capable model pulls the student model's capabilities up? What. Why the fuck is this even a question.
Look up literally any distillation works. Because this is just distillation but on one-hot token chains instead of richer logit KL proxies.
And no, I'm not claiming than you can "close the loop" and get RSI on the cheap just by distilling forever. I'm claiming that distillation is a very cheap way to bring the performance of a less capable model closer to that of a more capable model. It doesn't give you "a more capable model" out of thin air.
Which is why Chinese labs rely on Anthropic to provide that "more capable model" to them. They take the capabilities Anthropic trained for the hard way, and train for them the easy way.
It's a "fast follower"/"improved capability density" trick, not a "singularity tomorrow" trick. There are a few "distillation pump" tricks that get closer to what you have in mind, but they're still more about "extract more training signal out of the same set of data" than about "unbounded RSI".
Okay, you have no data nor evidence nor a paper backing this claim, it's just speculation.
You want to sell me the idea they are spending hundreds of millions to get unchecked Q/As with reasoning redacted and without checks on the output quality to do what exactly?
Have a shallow pointless bunch of expensive data to get slightly better RL? It's expensive and pointless.
Data has shown again and again that synthetic input/output does not benefit models in RL, it may even make the output worse.
Also, you have a giant bias.
The chinese are the only ones releasing models and research papers in the open from which American labs benefit 24/7 (DeepSeek has been copied by all US providers).
And you want to sell me this ridiculous idea of the giant return of spending hundreds of millions on unredacted pointless QAs?
What the fuck. Are you a literal, honest to god distillation denier? Straight up "wake up sheeple, model distillation isn't real"?
I've seen plenty of things in the dumpsters of AI discourse, but this got to be among the most baffling.
Yes, there are "giant returns" on distilling from a more capable model into a less capable model. And even more so when the more capable model was trained for something you want and lack. Like: better coding performance.
Someone like OpenAI had to RLVR for it the hard way (and if you think "distillation is expensive", wait till you hear how many bits per rollout hardcore RLVR gets you), but you get to peek into the results of their work and copy them for yourself.
Also, Anthropic didn't redact model reasoning until Mythos. OpenAI started with o1, but Claude had reasoning chains accessible for a long time. Which is why Anthropic was more targeted than OpenAI.
So we're meant to believe that only US companies have the intelligence and/or access to manpower to generate their own reasoning data? Does China have a population deficit? Maybe China has too high wages to pay people to generate reasoning data?
The US companies bootstrapped themselves from one model generation to the next, partly by using the previous generation to generate synthetic data, etc, and partly by paying people to hand generate training data for them. Why do you apparently assume that the Chinese can't do the exact same thing?!
Surely "coding performance" is by far the easiest thing to generate your own RLVF data for, since it has trivial verifiable rewards - does the code compile and do what you want.
RLVR is the poster child for model distillation. Because: have you considered just how many tokens does a model have to generate before you can check "does the code compile and do what you want"?
You generate 90000 tokens worth of rollout and get a verifiable reward once. RLVR is fucking expensive! It's worth it, because it often unlocks capability advances that other things don't. But it's still fucking expensive. RLVR eats compute like nothing else.
So, if someone used a lot of RLVR to improve a capability? Just distill from that "someone" and get a similar improvement for a fraction of the price! Then you can do your own RLVR from THAT cheap starting point, if you want to.
"Human domain experts" is a similar niche. Let's say hypothetical "EconomicsAI" hired some $200 per hour human economists to make training data for their "EconGPT" AI. What's cheaper - hiring your own $200 per hour economists, or using a bunch of "$10 per 1M tokens" outputs of EconGPT to bring your own model in line with what EconGPT can do?
Even synthetics can be expensive, because while synthetic tokens themselves are relatively cheap, the applied AI knowledge one needs to make high quality synthetics that improve task performance and don't backfire on you isn't. Again: distillation bypasses a lot of that - by cribbing from the outputs of a model someone has already done that for. Allowing you to get more oomph for cheaper, and spend your R&D effort elsewhere.
Your training cost argument makes no sense. It doesn't matter whether you are using human written code or someone else's LLM generated code to train on - you are going to be RL training on it, so your RL training cost is the same.
There is a data cost argument, especially if you are paying for human generated data, although I'm not sure how applicable that is to coding.
DeepSeek R1 was a famous case - not only it briefly beat then-SOTA on the cheap, it was also released with distilled versions that preserved bulk of the improvements but could be run on higher-end consumer hardware.
And of course Gemma models are said to be distillations of Gemini.
There are multiple stages of training, and the data/compute mix at each are quite different and produce different "layers" of intelligence.
The pretraining stage is the first stage which consists of "next token prediction" on the entire internet, PB of tokens, etc. This is what most people think of when they think of training LLMs, however it produces a "base model" which is not really "intelligent", but rather much like a blurry JPEG of all human language and knowledge. You cannot really talk to such a model; it will simply complete your prompt by producing both sides of the conversation. Note however at some level the training has encoded enough structure through compression that it is able to simulate all sorts of phenomena, from human conversations to code. The great R&D difficulty here is to scale pretraining so that it can proceed smoothly in vast distributed datacenters in a fault-tolerant manner.
The next few stages are collectively called post-training, and typically consist of supervised fine-tuning, then reinforcement learning.
In supervised fine-tuning, the model is further trained to predict the next token, but on a much more focused data set of natural language conversations where the "assistant" and "user" turns are explicitly delineated with special tokens. The output of this stage is a model which is capable of carrying on proper conversations, but typically with no ability to creatively problem-solve, and less of a personality. The data and compute are many orders of magnitude smaller than in pretraining.
The reinforcement learning stage used to be a small part of model training, but ever since AI-assisted coding took off, it has become larger and larger chunk of training. In recent models, the compute spend on RL has allegedly come to rival or even exceed that of pretraining [1], which is a bit scary because RL is classically what lead to sci-fi like AIs which are extremely good at accomplishing goals to the detriment of everything else.
The way that RL works is that you put an instance of your model in some environment (such as a VM containing a git repository) and give it a task (such as fix the linked github issue). The model will then generate a bunch of attempts to solve the task which we call "trajectories", in most cases there is either an objective measure of the task success (such as passing the tests), or a fuzzy measure (such as having another LLM look at the results and provide a score). This is called the reward, and the model will learn slowly by producing trajectories that receive reward. It can actually be quite hard to prevent "reward hacking" from the model here and the rewards must be shaped very carefully, much R&D labor goes into here, as well as similar challenges to distributed pretraining.
A significant challenge is that coding/knowledge work tasks these days are getting extremely difficult, we are far beyond 2024 days where models could barely solve the easiest problems in SWE-bench. Tasks at the frontier now look more like mini projects that would take humans multiple hours or even days to finish (or in some cases, research-style tasks that would be beyond reach for even top human experts, such as the Erdős unit distance problem which was posed in 1946 but wasn't solved until recently, by GPT-5.5). Huge amounts of trajectories must be produced, and huge amounts of them produce zero reward and therefore are useless for learning. Getting a cold start requires running tens of thousands of instances of your model in VMs in parallel for multiple days to produce trajectories, to say nothing of the GPU costs.
So what do you do when you only have a model which is capable of basic conversations but cannot even begin to tackle basic coding tasks, use tools, etc? The approach that companies behind the frontier have decided on is to bootstrap their learning process by having an already extremely intelligent model such as Claude produce hundreds of thousands of seed trajectories for them. Then they can use this data to get a warm start and begin learning immediately. And if you use Claude for your reward model too, you get to skip the nastiness of reward shaping.
Therefore, even if in number of raw tokens the data are much smaller than internet-scale pretraining data, the value that each token provides is far far greater.
Training isn’t a single homogeneous step. It starts with pretraining which requires bulk PB of data but you have less quality concerns here. You cover the whole data distribution. Later stages require less and less but increasingly higher quality and complex datasets. The late stage ones are highly curated and might even be sourced from world subject experts. This is where frontier labs with big pockets have the advantage.
From what I understand, at this point, the main value of stronger model outputs is simply to bootstrap reasoning behavior during the RL post-training phase. It gets you past the “cold start” problem with RL, after which the outputs aren’t needed anymore. From then on, it’s hill climbing and that requires environments for the model to interact with get rewards from.
> But if you show them a jailbreak of their model that bypasses their safety, they'll tell you that any model can eventually be jailbroken so don't worry about safety.
They claim two things:
1) The specific, available jailbreak for Fable 5 is not dangerous - this has been confirmed by multiple experts, and there is no credible evidence against this claim (in other words, Anthropic is probably correct)
2) It is impossible to build an LLM that is immune to all jailbreaks. Again, there is no credible evidence against this claim, i.e. Anthropic is again entirely correct.
If #1 was false, they could just publish the details of the jailbreak - it supposedly only works on Fable 5, so there's no possible danger.
If #2 was false, surely some other LLM lab would have done it by now. Especially since a number of governments have made it clear there is a market for such a project.
Supposedly the details of the ‘jailbreak’ are that you give it insecure code and say “fix this code”, and it does, and then you ask it for test scripts and that’s effectively an exploit against the unfixed code.
If that's the extent of the jailbreak, then the government should have banned every existing LLM - their story only makes sense if there's some Fable-specific capability that got unlocked.
> If #2 was false, surely some other LLM lab would have done it by now.
This is a logical flaw. LLM that is immune to jailbreak _could_ exist, but not yet, or maybe nobody talks about it. Yes there's a market, but all of these AI boom is too recent to make any claims.
I think pretty much parallel to how social engineering, manipulation, scams work. LLMs are being trained to have human values, prioritizing human lifes, yet people are shocked it will spurt out how to make a nuclear bomb because grandma is being tied to a train track as a hostage.
I would also spurt out how to make a nuclear bomb (ie public information you can find using google) if I was told that's what I gotta do to save grandma tied to a train track as a hostage. "AI safety" is such a shit show.
I'm guessing you mean, the incompleteness theorem guarantees that nobody can prove their model is un-break-able?
I don't think that's quite what it means. The theorem says that it's impossible to write a function, "will_halt(program, input)", that will be correct for all possible {program, input} pairs. But for a particular program, you may be able to write a proof that it will halt for all inputs -- that's what software verification is about.
The implications here would be that nobody can create a "will_jailbreak(model, input)" function which works for all model/input pairs. But we don't need a general function which works for all model/input pairs; we just need a way to prove that for a specific model, there will be no jailbreaks for any input. As with software verification, this may require that the model be developed in a specific way.
Granted we don't currently know how to make such a proof regarding neural networks; but that's not because of Gödel.
Similarly, if you did a corpus study on bioRvix to summarize recent science findings — you could use the same questions and answers to fine tune a model.
There is no way to communicate information at scale to companies through the API, for anything approaching a real application, without that information forming a corpus another model can be trained on.
But it wouldn’t be the first time they broke a model:
Their “guardrails” that cause it to reject user prompts also means it relies on its pop science summary of medicine to tell you why bioRxiv is wrong rather than accurately summarize the papers.
They’ve successfully created a smug, argumentative average of the internet which refuses to even consider it might be wrong or that it’s reading a science paper which is based on measurements and not vibes — but why would I pay for that?
Doesn't "real" distillation use the logits instead of the final tokens? I would classify this more like using a model to generate synthetic training data.
The compute deficit of Chinese Ai companies is real, and it IS THE ONLY competitive advantage that Western companies have.
The only way the U.S. keeps that edge is to prevent distillation.
The only way Chinese companies can make up for the deficit in compute is to distill.
There innovation in great supply on every side of the Ocean. Its about the chips.
And in terms of national security, for the U.S., and for China, its about the chips and the distillation that undermines that advantage.
This is an arms race.
If compute or access to training data were the only issues, then companies like Meta and X.ai (Grok) should be doing better, even Google for that matter. Musk even admitted that Grok used training data from OpenAI models.
While there is no moat as such, there is still a lot of expertise that goes into training SOTA models. There's a reason Google was willing to pay $2.7B just to get Noam Shazeer back to improve Gemini.
You got that wrong. The forcing function of compute scarcity is an advantage not a detriment. The amount of investment pulverized in performative model training and dead ends (Hi Sora) should make this obvious.
Good luck not crashing the markets and the economy.
And good luck not staying behind when you can't monetize your gargantuan investments and have little incentives to make your models better as the world moves on.
Putting aside agentic coding, that is to say, if you judge LLMs as a consumer technology (an old-fashioned idea for the inward-looking tech industry admittedly), then open weights LLMs, even quite small ones like Gemma 4, can likely already satisfy 90% of applications with a bit of help from search and browse tools.
Much of the arms race for better LLMs exists to satisfy only the IT industry's needs.
I've used RLAIF to build out heuristic based non-LLM models for various decision systems and achieved like, 95% F1 on certain projects. We're in a place where models can be used to fine tune a lot of stuff via loops.
> These complaints of distillation are inflating the problem to make it sound worse than it is
This is, in part, a problem every judicial and legislative system has faced since forever: form versus function.
Take a classic elicitation spying techniques: a foreign spy meets a military officer/scientist at a bar, strikes up a conversation, makes an observation wondering how could a missile hit some target at some accuracy and elicits a response that with laser guidance it is entirely possible. From this they get info that there is some technology to laser guide missiles. Or in retail, a competitor hiring a secret buyer for core baskets of goods and analyzing prices in the receipts.
The function is espionage, the form is conversation and all info is in a sense provided willingly. Where do you pull the slider?
These distillation "attacks" are not only indistinguishable from evals, they ARE evals. The function is own model training, the form is eval. Normally, one would expect to have risk benefit analysis based discussion which direction to push the legality slider to. The problem with these recurring statements is that they invoke enshitification of legislature.
I'm sorry, but you got the terminology exactly backwards. Training on the answer is called supervised fine-tuning.
Just for the sake of clarity:
0. Full distillation uses logits of the teacher model - that's much more information than the text itself. This is a kind of distillation used inside labs, but one can't distill Claude this way as logits are not available via API.
1. Supervised fine-tuning on synthetic data might be called blackbox distillation. I guess that's what you meant in your case (1).
2. Reinforcement learning (like RLAIF) uses least amount of information from the teacher, i.e. only few bits per task.
>But if you show them a jailbreak of their model that bypasses their safety, they'll tell you that any model can eventually be jailbroken so don't worry about safety.
Yes this is in line with what Anthropic said in their public statements about their Fable access restriction by the government directive. The hypocrisy and inconsistency in their statements and behavior feels quite childish and controlling. I believe our companies and their leaders, friends among our other influential leaders and leaders from rich social classes, want to actively hurt most people as this behavior looks to be quite self-interested.
Not to mention, the person who brought this quote unquote jailbreak to the Trump Administration was Amazon’s new CEO. They know their IPOs are coming up, so locking their competitors out of the U.S. (even if just for the weeks surrounding the IPO date) would be a major boon.
The White House seems to love making announcements just for the sake of making the market move…. Coincidentally, right after POTUS buys a massive amount of the benefactory company’s stock
(Buy Dell Computers, lol)
I'm not 100% sure it's not possible. If (I don't know) it's possible to freeze the temperature of the model so it's deterministic, and if you could make a map of produced words back to tokens (via HMM probably), then you can probably alter a minimal input and observe the output to model it. If you perform waves of such minimal alterations, you can expect to be able to locate the distance where each alteration impact the model (the idea being that a small alteration on output is likely due to the last layers of the models, and a small alteration is likely due to the deeper layer). Once you've located most of the last layer(s?) weights, you can try to solve for them. With a hundreds of billions weights model, the last layers will likely be so huge that it's probably unfeasible technically, but it's theoretically possible.
You can do things like that - one example is averaging weights between related models - but not with Anthropic's models, because outsiders don't have access to the weights.
> These complaints of distillation are inflating the problem
They’re also missing the point. What would have happened to a member of the Manhattan Project who, through personal pursuit of profit, neglected their duty enough to let the bomb leak?
The companies are all for-profit companies, its not like they're selling out some national security goal for profit, profit is the point.
Anthropic already heavily restricts Chinese traffic but that only jams up researchers and regular Joes. Anyone motivated enough can hop a flight to Singapore with an nvme drive in their pocket.
Chinese companies are engaging in anti-competitive practices, as usual. They are rogue actors on the economic scene. If it were feasible, they'd be widely banned, and for good reason.
Merely copying products that actual companies produce and making them cheaper is anti-competitive. There's no incentive for the products to be developed in the first place in a market if this is happening. This is why copy protections exist in civilized countries (not China and to a lesser extent India).
That's why IP was invented, but what IP are they infringing? Not patents, not copyright, not trademarks, so what? Making something cheaper than someone else isn't anti-competitive. There are a lot of businesses that do that. That's the very essence of competition.
Chinese resellers are offering Claude tokens at 70-90% below official Anthropic API prices. They achieve this by reselling capacity from pooled Claude Max accounts, payments fraud, and also reselling the model output & reasoning chains to various Chinese labs. They are subsidizing model access in exchange for user logs and reasoning traces, which they then sell as training data, allowing them to operate below cost.
Claude and ChatGPT are both blocked in China. You need to use a VPN to access either, and you can't pay with a Chinese bank card. So most people who want access to Claude buy access via a reseller. It's the easiest and cheapest way to access Anthropic models in China.
These resellers operate tens of thousands of bot accounts, which is also why Anthropic introduced identity verification, to slow down the onslaught of bots.
This is one reason why DeepSeek & GLM are priced so cheaply, they are competing with impossibly low token prices in China. They have to keep prices low, in order for people to use them.
> This is one reason why Deepseek & GLM are priced so cheaply, they are competing with impossibly low token prices in China. They have to keep prices low, in order for people to use them.
This one does not make sense to me at all.
Deepseek and GLM are openweights, even US inference provider are selling them at much cheaper price. The price is cheap because the model is more efficient.
DeepSeek permanently cut its V4-pro API prices by 75% because they were too expensive. Without the price cut, Deepseek V4-pro tokens would have cost more than resold Opus 4.8 tokens.
Opus 4.8 is a more capable model, so almost nobody was going to pay for V4-pro at the original price.
> Without the price cut, Deepseek V4-pro tokens would have cost more than resold Opus 4.8 tokens.
You mean it's functionally as if American tokens are being price dumped in China and Chinese model providers are being forced to compete with that and innovate? So many delicious layers of irony, lol :-P
Urm, no? I man they did cut prices by 75% that part is true - but they reduced a starting price that was below sonnet.
Also it's a open weight model, doing that is impossible long term because the real price will be set by the other model providers, who priced it around 60% of sonnet inference cost. Had to look that up though, so that's today's pricing.
Is there a contradiction here? If resold Opus tokens are sold at a 93% discount, you can be a lot cheaper than Sonnet while also a lot more expensive than resold Opus tokens.
I see, After rereading the comment I was responding to I realized I probably misread/misinterpreted what they wanted to convey.
I think there isn't a contradiction and I was just confused. The price may have been discounted only to get below the price point of opus resellers. I do not have enough information on that to make any clear determination on that topic.
It's somewhat difficult to have any sympathy for Anthropic here. They're entirely responsible for selling tokens at below cost, with the age-old bait-and-switch tactic.
If they weren't doing so, then these Chinese resellers wouldn't be viable.
Radical idea, but how about they actually charge a viable price, even on subscription plans?
If resold Anthropic tokens undercut even the at-cost open-weight model tokens, because they're reselling subsidized subscription tokens, then you'd have to start selling open-weight model tokens at a loss in order to match them.
>They achieve this by reselling capacity from pooled Claude Max 5x accounts, payments fraud, and also reselling the model output to various Chinese labs.
But is it cheaper than getting your own account? Otherwise this sounds like the "anthropic/openai are losing gazillions of dollars because they're selling $1k worth of tokens for $100" line that's commonly trotted out by AI bears.
It's very difficult for people to create personal Anthropic accounts from China. Anthropic blocks Chinese bank cards, so people must pay with a foreign bank card, which they likely don't have. And even if they manage to set one up, they have to access it via VPN, which eventually gets the account flagged. They then have to complete identity verification, which most Chinese users are unable to pass.
There's a similar Claude resale market going on in Russia. On Funpay they are selling Claude tokens for roughly 20-30x cheaper than official Anthropic API pricing.
China aren't offering a cheaper solution. They are subsidizing an existing one (which is already subsidized) in order to gain foothold. The difference is that in the US subsidies come from VC, while OP implies subsidies come from the AI labs that buy the training data (which may as well also be VC backed, so just one extra hop).
This isn't "the market working as intended", this is an exhaustion fight to the bottom where the one with most money gets to stay in the market. As with most venture capital startups. I believe this VC tactic is a well documented "cheat code" to bypass market forces and build a monopoly. I find it hard to compare that with a free market.
However, I don't really mind China "stealing" from Anthropic. For us consumers we are getting the cake and eating it too. I.e we are getting rapid improvement to the tune of over a hundred billion dollars in funding, yet the market remains big enough that there's a chance of it not ending up as a monopoly in 20 years. And venture capital are footing the bill. A part of their investment is practically being redirected to fund Chinese AI development. It lets us live out our lives as happy CAC farmers[1].
So I would argue its not as much of a "cheaper solution" as it is intentionally and maliciously abusing another company's product to extract more value than the billing plans intend (given an average user), and further subsidizing the product by selling this data to competitors. But I don't necessarily think its a bad thing for us end-users. Nor for the market. But it is bad for Anthropic and its investors.
> China aren't offering a cheaper solution. They are subsidizing an existing one
Chinese labs are also pursuing legit frontier-advancing R&D into efficiency and publishing papers in the open, a culture that's in retreat at top American AI labs
Their is plenty of innovation happening on both sides of the Pacific.
Again, China publishes open source because they don't have another game they can play.
They distill because they don't have the compute to compete.
They are great lab, for sure, but the fundamentals are driving their behavior.
The fact that are people that genuinely believe you can train an LLM by using random QAs obtained from another LLM is astonishing. Let alone the fact that it makes absolutely zero financial sense.
At this point this is being repeated so often that completely uninformed users are taking this at face value.
> China aren't offering a cheaper solution. They are subsidizing an existing one (which is already subsidized) in order to gain foothold.
In my economics classes, we were told that (in a "free market" argument) the best thing to do if a subsidy is making something you want cheaper is to use it. You're getting your thing, and at a reduced cost.
(I'm not really replying to you per se, I'm curious how "free market" folks in these comments would respond to this.)
This is why I don't understand why people complain about impractically cheap Chinese solar panels. The rest of the world should buy enormous quantities and bankrupt the mofos and hugely benefit along the way. Then later they can set up their own solar panel industries.
Because they arent selling at a loss. The business pipeline is subsidized by the state. But end to end from mining the minerals to shipping you the solar panel everything is "in house". Its all in China. Thats why why can sell so cheap. Its even cheaper to make.
This narative that the CCP is just subsidizing all business to "beat america" is just dumb. Its the build process being made cheaper by the government. Not the final product.
Let’s not act like the US government isn’t subsidizing AI either with massive contracts. Anthropic is selling subscriptions at a loss; reselling tokens is just arbitrage.
I’m sad for you that propaganda has destroyed your critical thinking ability. Qwen is an Alibaba product, Deepseek is from a private Chinese hedge fund. Those are not the CCP. The Chinese economy has vastly evolved in the last decades. The CCP doesn’t have any more special control over the Chinese labs than the US has over our labs. The White House can do whatever it wants to keep labs in line. Fable just got pulled because the US gov ordered Anthropic not to give access to any foreign nationals, including Anthropic employees.
Please, show me one Chinese court ruling against the CCP and I will believe you. Anthropic can go to court and have the order overturned if it isn't legal (with legality being born from elected representatives), it happens all the time.
Just because Xi Jinping lets companies play mock "private businesses", does not mean there actually is private business. At the end of the day, the CCP still has final say in everything, and Xi has final say in the party. There is no constitution (in the US judges swear to the constitution, in China they swear to the party), and there is no balance of powers.
It's just one guy, running experiments the way he see's fit.
Well they shot down his landmark tariffs for one, and the next court ordered refunds so...I'd be delighted if you could share cases of the government ignoring court orders (not to be confused with challenging them, like any functioning legal system has).
Also, obviously Xi doesn't make every decision. No dictator ever did that because it's impossible to do. The distinction is that no one has ever (or has the ability) to over rule what Xi decides. So if Xi has a stroke and wants DeepSeek to start manufacturing underwear, they will be ordering sewing machines tomorrow. Any sense of "private" is a farce.
As far as I can tell, the Chinese government itself is complaining about 'oversupply' in the solar panel market. Ie it doesn't sound like they are subsidising it anymore.
The Chinese government has stopped direct subsidizing solar panels years ago. I think it was around 2019? This resulted in a lot of companies going under at the time.
It did not stop solar panels getting cheaper and cheaper because of the whole integration and mass production (with healthy free market competition).
The last subsidies like export value-added tax rebates for solar panels and lower rebates for batteries are ending in 2027.
China their main power is, the ability to have everything inhouse. Yea, they subsidize a lot of stuff until it hits critical mass, and then you have often a healthy industry with lots of competition.
China alone has like a few 100 car manufactures because of the subsidies, and over time there will be consolidation / buyouts etc but the end result is a healthy new industry that exports. With again, everything internally being produced.
This is why our subsidies fail. We do one sector, often a few companies at best. This results in few competitors, expensive prices, and often reliance on externals that can bankrupt those companies. And que how we wasted again dozens of billions in propping up a industry with no competitive edge.
People can cry about China but they are actually doing work, despite the mass amount of corruption. That is the big difference with here... Mass corruption got in the way of national security, plop, people go to jail. Industry quickly gets their ** together. Here ... give billions, and the money vanishes, with no real consequences.
Local governments are over-funding numerous producers (though cheap loans and other subsidies and incentives) creating excess competition. This is an ongoing problem and is a huge misallocation of capital. Increasing demand just drives this process harder and puts downward pressure on margins. As soon as they try raising prices, or just through satisfying total demand, demand collapses and they (almost) all go out of business.
The Chinese model has weaknesses, we should be exploiting them.
the fact that you think an organization that pulled 300 million people out of poverty in 20 years with strategic planning and a controlled economy has this not covered is mind blowing. they killed the made in USA slogan in less than 40 years. they'll be fine.
The textbook poverty which are created by such organization itself with strategic planning and a controlled economy in the first place, killing ~30 million. All the more impressive it only took them under 10 years.
> China aren't offering a cheaper solution. They are subsidizing an existing one
So basically like US companies subsidizing offerings with selling user data, ads for crypto scams, manipulation for elections, making people addicted to gambling and so on?
Seems fair and an improvement as you can choose between that and not. Unlike say offerings from Meta where the data selling and efforts to further gambling addiction is always included.
Which part are we supposed to have an issue with?
The selling data to offer cheaper compute?
Products taking over markets with below cost pricing because they have money and ruining the free market?
Because all of that is considered totally okay when every single US big tech company does it.
All I can say is lol. DeepSeek showing 3 order of magnitude efficiency gains over the performative capital furnace that was training and inference absolutely moved the bar here.
Chinese models are years ahead of american models on multimodal comprehension, better yet,they publish on what makes the models tick and release weights openly.
Chinese research outout, publically released, has also contributed in big ways to features present in every single US model. Yours is a bit of an unfair take I'd say.
Besides, claude will think its chatgpt sometimes, so clearly this isn't a problem restricted to china, turns out unethical companies will do unethical things /shrug
> This isn't "the market working as intended", this is an exhaustion fight to the bottom where the one with most money gets to stay in the market. As with most venture capital startups. I believe this VC tactic is a well documented "cheat code" to bypass market forces and build a monopoly. I find it hard to compare that with a free market.
Why? Lots of people try this tactic, but hardly anyone ever succeeds. Meanwhile, the customer benefits.
> This isn't "the market working as intended", this is an exhaustion fight to the bottom where the one with most money gets to stay in the market.
That's, uh, pretty much exactly how oligopolistic markets function.
> I find it hard to compare that with a free market.
Well, to have free market you need to remove as much barriers to enter the market as possible. Huge capital investments required for entry and intellectual property laws are two examples of such barriers. Subsidies kinda supposed to help alleviate the first one.
I mean, for what it's worth, we have subsidized Anthropic by allowing them to train on copyrighted stuff. (I know it is still legal, and I support the legality, but the economics are what they are with people's content paying a big one time subsidized cost (to the level of at least 500B).
I am mostly economics illiterate but I understand a subsidy to be an economic concession given by the state to an entity which gives said entity a relative advantage compared to its peers.
In that sense (which could very well be bogus), letting a company violate individual IP of basically every human is less of an economic concession and more of unconsented to IP open season.
Even if one were to drop "economic" from "economic concession" and instead view a subsidy through the lens of a more general concession, one could say that the US Govt gave US AI companies a legal concession to sidestep the copyright protections of other US entities. But the US Govt should only get to undermine the copyright protection of other US entities - who gave American companies the right to violate the copyright of non-Americans?
That's some "download a car", $100000 per infringement pricing logic. No one is paying anyone 500 billion dollars. I'm sure rights owners wanted that, and more too, but it's nonsense to call it a subsidy that they didn't get it.
If we as individuals were sued it surely would be at least an order of magnitude difference between what is required from us vs Anthropic or OpenAI. That’s even completely ignoring the marginal utility of money. It is absolutely a subsidy. It’s just less fair because that power, to pay pennies on the dollar, is only given to corporations.
The VCs footing the bill is really your pension funds and 401Ks and banks passing through the VCs. If VCs lose money the contagion spreads through the economy.
Seriously AI companies complaining about fair use is the biggest case of crocodile tears I can think of. Irony has been dead for a while, but they dug up the corpse and set it on fire anyway.
I would assume China is working on liberating Anthropic weights through the battle-tested strategy of finding someone in a privileged position and getting them laid, etc.
black and white hat is relative. someone breaking into a state run database in a dictatorship and stealing documents that prove some opposition leader was murdered would be a black hat criminal if you ask their government. a hacker jailbreaking a phone to let people fix it without expensive official service is a black hat to the company. we should really switch to saying offensive and defensive or something else that doesnt come with moral implications. maybe lawful and chaotic.
Ok, but what about those shady sites that resell Windows education keys? They're certainly a "better experience" than buying legit keys, by virtue of being significantly cheaper. You aren't even really committing copyright infringement in the process, because Microsoft gives out windows isos for free, and the seller is really selling a random 25 character string, which can hardly be copyrighted.
>If there’s any bootlegging going on it’s Anthropic that’s doing the bootlegging but having mirrored the video etc sufficiently to beat copyright law.
>>If there’s any bootlegging going on it’s Anthropic that’s doing the bootlegging but having mirrored the video etc sufficiently to beat copyright law.
>US courts have consistently ruled it's fair use.
And they also have ruled that the that output of an AI isn't copyrightable.
As such copying claudes output isnt even fair use as that is an exemption to copyright but the same as copying public domain work which any and all are allowed to do.
> because Microsoft gives out windows isos for free,
… with a license that only allows you to use it for certain purposes, subject to certain restrictions.
> and the seller is really selling a random 25 character string, which can hardly be copyrighted.
1. Copyright is about creative works. It is possible to have a meaningful creative work no more than 25 characters long (or equivalent). Music is particularly good at this.
2. The key itself is not copyrighted (it’s not a creative work), but is reasonably interpreted as a copyright circumvention device. See also https://en.wikipedia.org/wiki/Illegal_number.
> Ok, but what about those shady sites that resell Windows education keys?
Yes, they are fine? They might no longer include full first party support by Microsoft for not being "new". Same as buying a used car (also comes with the "shady sites" for a far longer time).
Though this not making any difference by Microsoft not doing any support either way to make more money is a business decision by Microsoft.
What about those store brand cereals? “Chocolate puffed balls” for a fraction of the price of Cocoa Puffs™?! You all may laugh until your waterways are under siege and you find Cap’n Crunch™ keelhauled by thrifty shoppers
The current case law in the US is that the raw output of an LLM cannot be copyrighted without further meaningful arrangement or alteration by a human author.
Free market would of course allow bootleg DVD sales, state regulation that gives monopoly rights restrict it.
In the context of LLMs, monopoly rights haven't been created (yet anyway).
Fun fact: for a period the US (or american colonies) didn't have copyright but Europe did, so people could copy and sell English (and other) books for free.
Maybe the making of the thing should be paid for before it's made, rather than hoping that selling copies will recoup the investment. I.e., go back to patronage while abolishing copyright.
This type of "resource curse" paints a perfect picture of why US based frontier providers are set up to fail. They want, and have, few restrictions and along with that unlimited warchest. The Chinese on the other hand aren't burning billions like millions. Anthropic, OAI, Google, Meta... They're all phenomenal examples of waste, corporate greed, inefficiency and are the reasons people hate tech bros at this point. Whining and crying from their super yacht, Parkinson's law is alive and well!
any third party provider can offer zdr. if its a reputable company in a place like switzerland or germany i would trust them more than anthropic to hold up that promise.
> Yeah, like all those Chinese bootleggers selling DVDs for a few dollars rather than $20. Free market!
It's supremely ironic analogize distillation to copyright infringement when it's literally what Anthropic was found guilty of. It's not illegal to distill. It is illegal to pirate. And it's what Anthropic was found guilty of, not Alibaba.
I get the vague impression that this was written in a sarcastic way, but it has a straightforwardly true literal read because yes, this is what the free market is about and Anthropic will have to compete with the Chinese if they want a big share of the market. Chinese models are cheap and good; even without reselling Anthropic's services they're competitive. Which reading did you intend?
And, gotta say, the idea that the Chinese are better at selling US models than the Americans is hilarious. There might be an economic study here somewhere about just how anti-consumer and anti-progress their IP laws turned out to be. We've got an entire postindustrial revolution centred around who can ignore the most stupid laws.
Given the current US government's tightening of export control restrictions and the introduction of a bipartisan bill to block use of Chinese AI in federal agencies, I'd say the two countries' positions are not far apart.
That is ALSO happening, but that's beside the point.
Chinese AI apps like DeepSeek are freely available for ordinary Americans to download and use. There's no federal law banning private citizens from using them.
So to claim that Chinese companies are better at selling American companies' work than the American companies can do themselves when they are prohibited from operating in that market, is the wrong deduction to make.
What is the implication here? Are you warning that US corporations might start doing something shady, like scraping the internet at large scale for training data? Or mass-dowloading pirated copies of books, completely ignoring copyright?
I find it hard to imagine a future where US corporations have degraded to such a point.
No, he means that the US will close most of its domestic market to competition just like China has for decades, and the US may start subsidizing and dumping its goods everywhere
> the US may start subsidizing and dumping its goods everywhere
Isn't that exactly what companies like Uber have already been doing? Take VC money, sell goods & services at a huge loss, wait until the competition goes bankrupt.
Exactly, it's funny how most Americans have no self-awareness on this topic.
And beyond VCs, which are like massive subsidies funded by printed dollars to which no other country has access, even in industries like electric vehicles, Chinese total direct subsidies to their EV companies are like $5bn per year, while the the ones provided by the US to their auto manufacturers are in the range of $50bn per year.
I don't think the US are cheaters or are doing something bad. But i do think that this propaganda about China flooding the market through "overcapacity" and subsidies is very dishonest and needs to stop.
Yes. Dumping abroad is the entire model that Silicon Valley has been built on in the last 2 decades. China just copied the model. And even then it's a light version of it.
What’s worse, tariffs or outright banning the competition from your market? China has done both despite globalism being what has lifted it from poverty. Why is everyone suddenly surprised that globalism and free markets are coming to an end? Is this a net good? Mostly no unless you count more redundant supply chains.
> China has done both despite globalism being what has lifted it from poverty.
such oversimplification on steroids is totally misleading.
globalism was never invented or promoted to help any country in poverty, it was designed to extract excessive values from those poor countries in the first place. painting globalism as something noble is naive at best.
globalism was the theme of world trading for the past several decades, it was available to all nations. care to explain why other nations in poverty failed to be lifted by the exact same fancy globalism?
let me help you on this one - China was THE leading technological and economical force of the vast majority part of human civilisation. What happened between 1840 and 2010 (the China in poverty period) was an outlier of the history. Globalism didn't lift China from that poverty, the ability to lead the human civilization which was embedded into the Chinese DNA did that.
Kid, when our Chinese ancestors wrote the Art of War, your ancestors were still swinging on trees. You just missed that big picture.
> such oversimplification on steroids is totally misleading.
Yes, so the kettle is calling out the pot?
> globalism was never invented or promoted to help any country in poverty
It doesn’t matter what it was designed for. What matters is what it does in reality and there is no doubt that globalism helped lift China from Mao’s disastrous policies. That’s not mutually exclusive from China’s past as the Middle Kingdom
To steel man the person you are replying to, what reduced poverty in China was money from globalism + significant domestic reinvestment of that money into poverty reduction. That reinvestment policy was a deliberate choice, and now China has the biggest middle class in the world.
An example of a country which didn’t do that is Nigeria. They got something like $300B in oil revenue over a 30 year period but have actually seen significant increases in poverty, now at 70%.
The surviving non-American farmers would be confused by the future-speculative tense as America has already been doing this for decades in agriculture, and have been complaining for decades about both the subsidies and dumping of American corn.
Sure, but not at China’s scale and no where close to number of industries where China does it. Why? The US was a net importer in order to support the dollar being the global reserve.
Whole Silicon Valley is based on selling products under price, for years, killing the competition or making it impossible and extracting once monopoly position is stable enough. It is the same play book again and again and again and again. It runs unprofitable companies for absurd lengths of time.
Most of the Chinese domestic market is open to foreign competition. The areas that are closed off are those that are politically sensitive: publishing (including social media) and banking.
As for dumping, Chinese goods generally sell at a markup abroad, which is the opposite of dumping. Chinese tokens cost more abroad. Chinese cars cost several times more in Western markets than in China.
"Dumping" is when Chinese companies beat Western ones on the free market. If all claims of Chinese government subsidies on basic products were true, China would've gone bankrupt multiple times already.
You're being beaten by a Chinese company? Why improve your own process when you can just lobby for sanctions and tariffs instead!
At least in the case of solar and EVs, it's a case of western countries preferring to protect their existing cashcow industries rather than invest to build the industries of the future.
For a brief second, Germany was in a position to become a solar power global player. But our conservative government was more interested in protecting their local, bad industry. Including destroying forests for coal all projections said we would never actually need.
From a EU perspective similar could be said about the US market - no strict worker protections, lobbying, and a general "capital first" mindset over the users/people (see GDPR etc).
That does not explain DeepSeek, nor does it explain the car industry.
The main advantages the Chinese car industry has right now are: they lead in battery R&D, production is highly automated, they iterate quickly, Chinese work culture is extremely competitive and things get done fast, and the Chinese state has policies to promote EV adoption, so there's a huge domestic market.
Note that the last point is different from subsidies to car manufacturers. Cities made it difficult to get license plates for ICE cars. The government encouraged the massive buildout of charging infrastructure. And it used consumer rebates, like California did.
aside from the huge domestic market (or potential in the future), china has built incredibly efficient infrastructure for manufacturing prototyping/production.
but it's also thanks to protectionism, and their strictly controlled (not freely traded) cheap currency.
if china had to play by the same rules as japan or germany it would not be quite as successful. but the west walked into this trap, hoping their win-win proposal would be satisfactory for all. now the west is too dependent on chinese production to enforce equal standing.
of course the US has its own unfair advantages, e.g. the global reserve currency and the massive post-WWII headstart.
The US spent decades transferring manufacturing, capital, and know-how to China, while Chinese students trained, and excelled, at elite Western universities. Why are people surprised that China eventually became capable of competing with the US?
Hostile spy agencies are now as focused on infiltrating western universities and companies as they are on doing so to governments, according to the former head of Canada’s intelligence service.
David Vigneault warned that a recent “industrial-scale” attempt by China to steal new technologies showed the need for increased vigilance from academics.
“The frontline has moved, from being focused on government information to private sector innovation, research innovation and universities,” he told the Guardian in his first interview since leaving the Canadian Security Intelligence Service (CSIS), which is part of the “Five Eyes” intelligence sharing alliance with the US, UK, Australia and New Zealand.
People like Mr. Vigneault know nothing about how academia works. If they get their way, they'll do massive damage to Canada's academic research ecosystem. Academia is naturally open and international.
These people don't get that academics publish their research in openly available journals. They go to conferences around the world and tell everyone who will listen exactly what they are working on. Unless you're working in a secretive government weapons lab, there's nothing to hide.
In the US, people like Mr. Vigneault instituted a witch hunt against ethnically Chinese researchers, and ended up messing with the lives of all sorts of innocent people, including the director of MIT's mechanical engineering department. They found zero spies. Just a bunch of scientists working normally.
> Chinese goods generally sell at a markup abroad, which is the opposite of dumping
Dumping is selling goods below cost.
Usually because government is subsidizing part of the production. I don’t believe the word “dumping” is used for the similar process when Venture Capital is subsidizing it, but using the same term would make sense.
Price at home vs. abroad is key. The term dumping comes from the idea that a company that sells profitably in its home market dumps excess production abroad at below cost.
This is not what is happening here. Chinese manufacturers are making a large profit off every car they sell in Western markets. As I said above, they're selling these cars at several times the price they charge in China. Unless you believe these cars are being sold at just 30% of cost in China, there's no way Chinese companies are selling below cost in the West.
> I don’t believe the word “dumping” is used for the similar process when Venture Capital is subsidizing it,
I've been doing so for years. How about you join me today. I already see two other users doing the same, so there'll be at least 4 of us.
It's blatantly dumping, whether the source of the money is directly the government (those in power) or VC (mostly US billionaires (trillionaires?), in other words, those in power) is a trivial implementation detail.
In debt the first 5000 years Geaeber makes the case that pure “free market” trade has never really existed in “the west”. The closest to this ideal that’s ever happened was during the Islamic golden age enabled by religious prescriptions against usury.
>The closest to this ideal that’s ever happened was during the Islamic golden age enabled by religious prescriptions against usury.
How does are bans against consensual financial exchanges close to the "ideal" of the free market? It just sounds like you have an axe to grind about the financial system rather than describing free markets.
What makes this view more correct than say, "economies with marketing creates a dynamic where being competitive in production is secondary to marketing" and concluding that nothings a free market until we ban all advertising? After all, you can make a vaguely plausible argument about how marketing isn't really about the merits of the product, and therefore allowing it is antithetical to the free market or whatever
> What makes this view more correct than say, "economies with marketing creates a dynamic where being competitive in production is secondary to marketing" and concluding that nothings a free market until we ban all advertising? After all, you can make a vaguely plausible argument about how marketing isn't really about the merits of the product, and therefore allowing it is antithetical to the free market or whatever
Wait, so your pitch in favor of a debt-fueled market economy is that advertising is awesome and that we wouldn't want to "lose" being smothered in ads all the time?
Cause... sign me up for the non-financialized, non-mass-media-advertising-driven economy please and thank you. I'd even be ok with just nuking billboards and mass-media forms of ads and still allowing more direct forms of marketing, if we must compromise! Likely we could find some compromises around just how much of the debt world we regulate too (this should be obvious?).
(I thought the disconnect between the efficiency of competition and the market as realized in modern economies was pretty well understood and taken for granted, but I guess we all find ways to justify the system we're profiting from... even if that means we have to claim we love the ad breaks)
>Wait, so your pitch in favor of a debt-fueled market economy is that advertising is awesome and that we wouldn't want to "lose" being smothered in ads all the time?
The point is that if add random caveats to what counts as free market, it won't be "free market", only "market I like".
I'm flabbergasted that you look at the Chinese property crisis and say "only the West does irresponsible loans." No, 60% of China's economy is state-run companies and the remaining 40% need political officers. China is just as capable of making shortsighted decisions as the US, and they have already made several devastating ones.
>I'm flabbergasted that you look at the Chinese property crisis and say "only the West does irresponsible loans." No, 60% of China's economy is state-run companies and the remaining 40% need political officers. China is just as capable of making shortsighted decisions as the US, and they have already made several devastating ones.
While these are hardly shy claims, I don't see anything in them to say "only the West does irresponsible loans"?
> The West is in a state of psychosis with Debt and Monopolies under the illusion of free market.
> The Chinese markets are more free than West, you can just look at the Auto and AI industry.
or the prior post
>Usury and debt based economy creates a dynamic where being competitive in production is secondary to financialistion.
> In short, instead of market being driven by demand and productivity, it is driven by financier curving out monopolies.
> Peak Examples are Uber and AirBnB.
You can throw a rock these days and find a category where the products coming out of China are miles ahead of those coming out of the rest of the world, from a bunch of companies nobody had heard of a few years earlier. And the list is growing pretty steadily.
I would assume plenty of shortsighted decisions are also being made. But I would have a hard time characterizing the state of competition in the west as healthier or more productive when looking at the number of players and the quality of goods being produced in China.
Capital will not be hoarded and stuffed in pillows without Usury. People are happy to take bets for profit and loses, I mean, that is like the entire stock market schtick.
...except Uber STILL faces competition, and I went back to hotels after finding AirBnB too pricy.
It is good and proper that people aim to create monopolies, as long as they want to do that in a productive and legal way! Monopolies are inherently dangerous, but the truth is that acquiring and maintaining one is not straightforward unless you can get the government to ban your competitors.
a stranger is asking you to risk $100k on his half-baked plan in exchange for nothing, and you say "sure go ahead take my money!"? no. it doesn't work like that in the islamic world.
as a borrower who's not allowed to compensate for your lenders' risk monetarily, your access to loans is severely restricted. Essentially you have to rely on your extended family. and instead of paying for the risk with interest payments, you have to pay with loyalty and subservience.
it restricts social mobility far more than the western model. it incentivizes clan structures. which incentivize cousin marriage.
power concentrates in the patriarchs of a million little family kingdoms. which causes all kinds of economic inefficiencies.
in the US, even if you're born without any family connections, as a healthy 20 year old you can find a job (hard work) that allows you to save $70k per year and invest it. when you're 30 you have $1M and a good credit history, you can easily leverage that to get a $2M loan at low interest rates, which allows you to start any kind of productive venture you want.
and you can do all this without owing your clan's patriarch access to e.g. your most profitable clients, or your daughters hand in marriage to his retarded son, or anything else he wants in exchange for his generosity.
All interest is usury in Islamic Law (and the laws before it such as Christianity and Judaism). There are ways to to put free cash into productive use without exploitation.
That's not true. Islamic finance forbids indefinitely growing interest. Sharia finance agreements involve fixed fees or equity shares. Late penalties can be collected but must be donated, not profit. In all cases, the borrower never owes to the lender for the lender to keep more an a fixed amount determined at the strat.
AI was always going to be a race to the bottom and low margins. It’s why I’m extremely bearish on AI as an investment. It’s framed as some high margin business when it’s really going to end up like your toilet paper at Costco. You will use whatever is cheapest and gets the job done.
And the value-add experiences that utilise LLMs require immense imagination et al that folks at Anthropic will not be able to conceive of - given that they have made immense sunk investments in existing assets. This clouds ones thinking immensely.
Both OAI and Anthropic have tremendous failure risk and this is of course not reflected in the fake private market valuations.
I see a world where lots of stuff is mass produced in china (tokens) but the acutal goods that deliver the experiences are designed, marketed and sold in the west at much higher prices. of course this a nightmare scenario for anthropic et al.
I used to think this.. but I think my opinion is changing. The reason is that the leaders likely will be able to accelerate faster.
So what you see is the market "stretching".. the bottom getting cheaper and the top end running away and getting more expensive. At some point the top end may be too valuable to even sell access to.
Most white-collar/knowledge-service-industry work is a weird type of work.
It's fundamentally about enabling things and largely middleman-type stuff. I have a hard time imaging what "At some point the top end may be too valuable to even sell access to." would even look like? What are you doing with that AI power, and who is paying for the output and why?
Elon probably isn't gonna spend that much on a model that can generate him ever-better fake porn but does nothing that he can use to sell stuff to other people. Especially in a world where open models are "good enough" for many things like "tell me how to fix the plants in my garden that are dying" and the like. What remains in the narrow knowledge-work space of: can't be done by an individual or small group themselves, but is valuable enough that it would make sense for people to hoard access to these extreme frontier models? Try to recreate Hollywood-as-a-monopoly by becoming the single content producer for everyone's individualized feed and so owning all the advertising budget in the world? Seems hard, we've already seen how easy it is for cheap-and-crappy-but-addictive-or-funny content to disrupt traditional media.
(There's also pure scientific research, but historically that's not very directly connected to "massive profit" and has a habit of leaking out and getting productized most effectively by other people or just being really easy to copy once someone shows how it's done.)
Robotics could be a different story, as physical labor can be more inherently productive, but "reasoning" advantages are unlikely to be a big long-term differentiator there. At some point the brick laying robot is satisfactorily building the structure, and you're good.
A huge amount of the value of "the economy" and the power of a currency is driven by circulation of money, and one thing that all the "bullshit jobs" white-collar/service-industry work does is keep the money moving and ensure that a lot of people have some good-or-services of value to exchange. If you take away the ability to offer services worth exchange from huge chunks of the economy in these super-frontier-models-replace-everything scenarios... you're gonna have a bad time?
> The reason is that the leaders likely will be able to accelerate faster
Model improvement is already hitting diminishing returns, and people aren't willing to pay substantially more for a slightly better model. There's no "accelerating away" when the new models don't open up a huge new market. If anything, the companies burning huge amounts of money on marginal improvements will be undercut by companies happy to sell current models at a significantly lower cost.
Glm 5.2 very much argues against that. Opus 4.8 level quality for cheap. That’s sufficient for most tasks, so if/when you do need SOTA models you can spend more for specific tasks but otherwise rely on the cheap but still plenty good models for everything else
For me better models are like 8k TVs, my 4k TV is fine, I really don’t know if I can see or tell the difference from 8k and 4k, and to be honest I’m usually just streaming some 1080p anyway. Sometimes products reach the plateau where humans just don’t need better. I’d certainly never pay for AI, Gemini 3.5 Flash free works just fine when I need AI. I don’t even click higher models for free in the Gemini app. I mainly just care about speed. I’m not a programmer, AI doesn’t make me able to make my job better or make more money and the vast amount of people in the world are like me. These valuations are not based on that reality and the stock market correction is going to be 1930s level I fear.
Free markets work when paired with property laws that can be enforced if broken. If China could offer a cheaper solution in that framework, it would be as you say.
If you keep studying econ you will learn that these failures are actually the norm, and thus why the only "capitalist" states to really succeed have been the ones where the state was strong enough to reign in the market.
Of course, such a state of affairs is temporary at best -- since the alternative is so lucrative!
The "free market" gave the PRC its current strategic lock on rare-earth minerals. There's definitely no such thing as a free market in a Maoist dictatorship. I personally think the "free market" concept is an unachievable ideal and thought-terminating cliche, but "free market in a Maoist dictatorship" is for sure a contradiction in terms.
If you can use mountainous capital to sell at a loss in order to distort the market, yes, that's not a "Free Market", as in, the vaguely understood competitive marketplace armchair economists idealize.
True freedom in the market means the freedom to capitulate your wealth to snake oil salesman and schemers who operate on generational timeframes until economic power consolidates and renders your society into de-facto tyranny. Before any sort of regulations existed, we were all trading shiny rocks with ultimate freedom, and that somehow has produced a bunch of economic situations in the modern day that a ton of people don't like.
What's more interesting to me is freedom from the need to have investigative journalists doing deep dives into potentially fraudulent, thieving, or scheming companies behind every purchase, and to know that what I'm granting market success to is exactly what my money or time is going towards - I'm not buying something at a loss that funds some other deliberately obfuscated project that's made opaque from my perspective of the market transaction.
The proverbial "market wisdom" doesn't emerge out of markets with extreme information asymmetry.
Externally subsidized predatory pricing is the opposite of a free market — precisely because it sells things at below market rates.
Free markets are where players compete on quality, efficiency, and supply. Prices are a result of cost and supply and provide real information on these factors. Competition for customers selects the most effective and efficient producer.
Sustained efforts of selling at a loss to gain market share is the exact opposite. The entire purpose is to corrupt the free market by sending false price signals which SUPPRESS free market competition and push market share to whoever can burn the most capital (whilst providing an actual service/product), not whoever is most efficient or highest quality or lowest actual price provider.
Uber and AirBnB are better examples of your "selling at a loss to gain market share", where they burned capital to undercut prices for close to a decade on falsely low pricing to destroy incumbents.
Spending on R&D while developing expensive technology is different and arguably very much a part of a free market, and is not what I was talking about.
Spending capital to steal your competitors' technology, and then spending more of it to make it available at below-market rates, is absolutely not a free-market activity.
Just because it is not stopped by someone enforcing a free market, does not make it a free market.
2- getting banned and creating thousands of accounts to break the conditions of the service at scale
3- using VPNs and proxies (possibly residential) to mask their network origin and identity
4- Using potentially fake names to sign up
5- Using different credit cards?
Fraud on so many levels, a lot of the infrastructure and modus operandi is what cybercriminals use, these are attackers man, whether you like the victim or not, and whether you think it's poetic or not, I recommend compartimentalizing and just trying to gauge whether an act is wrong or not in itself.
Do you also think Chinese selling counterfeit US postage stamps on eBay for 50% retail price (which is a major problem CBP and USPIS are fighting presently) is the free market at work?
This post is so delusional and dripping with condescension I've read it three times and I still can't figure out if you're trolling or not.
If I understand your argument it's ethically ok to destill huge swathes of copyrighted work into a model without compensation, but then it is ethically wrong to use that model without compensation (well actually reduced pricing)?
I don't get the moral framework that you're applying. Could you elaborate?
Over the air TV also isn’t public domain. It’s licensed to a station for broadcast. The output of an LLM has been deemed ineligible for copyright. Until you square that pickle your circle isn’t circling.
Free over-the-air network TV is (generally) copyrighted.
The output of LLMs cannot be copyrighted. This isn't a semantic game; it's literally the case that Anthropic cannot seek relief for people duplicating the output of an LLM.
The relief available to a licensor for violating a license use restriction is cancellation of the license. And they're free to do that, just like Alibaba is free to pay somebody in Hyderabad $20 to make another one.
DMCA can't apply in this case because (this is the "C" in its initialism) it is based on copyright protections, which the output of Claude is not eligible for.
I won’t go too far into the weeds, because I’m not an internet lawyer, and I basically agree with you, but I do believe there are access restriction laws that are not only limited to copyright violation. People have gone to prison for enumerating sequential identifiers in URLs to access records they shouldn’t be able to. I don’t know if Anthropic could actually make a case there, but it seems plausible at least.
Using a bunch of nonsensical/irrelevant analogies to somehow make a point seems worse than these “word games”? What does streaming copyrighted content have to do with LLM outputs (which are public domain)?
No, I mean that dodgy Chinese firms are cheating their customers:
> Because users’ inputs and model outputs are mediated through a proxy, users cannot verify which model their request was actually routed to. A user selects Opus 4.7, but the proxy can silently route to Sonnet, Haiku, or, in the worst case, GLM or Qwen, and fraudulently relabel the output. In a recent paper from Germany’s CISPA Helmholtz Center for Information Security (which cited my article last year on grey market!), researchers audited 17 API proxies and found widespread model swapping–API proxy access to “Gemini-2.5” achieved only 37.00% on a medical benchmark, a staggering drop from the 83.82% performance of the official API. On the user end, the tell only comes on complex tasks, when the output feels off (often referred to as 降智, or “dumbed-down”), but there is no clean way to prove it. Numerous public records highlight concerns that certain API proxies have noticeably compromised model performance. These proxies are suspected of “diluting” (掺水) services by substituting premium frontier models with inferior tiers.
> Besides model swapping, overconsumption of tokens also makes the price per token cheaper, though at the expense of driving up the total cost. Some of it is structural, as proxies that rotate accounts frequently destroy cache continuity as a side effect, forcing users to burn full-price tokens on context that would otherwise be nearly free. Some of it may be deliberate as the proxy providers try to milk more usage. The line between the two is difficult to draw from the outside.
I mean, which lawyer caste do you respect? Is that one is cool with stealing credit cards to buy Claude subscriptions?
> 3. At an Italian airport: Constantly stealing bags, opening them to pick out MacBooks and credit cards, a credit card manufacturer-who sells stolen "black" credit card info to transfer stations— is racking his brains to save you money.
On the one hand they talk it up as world ending and on the other hand they can't manage bot accounts on their own service.
I want to hear how this can be rationalised.
From the article "every layer of control frontier US AI companies have added (geoblocking, phone verification, credit card requirements, and now live biometric KYC checks) has produced a corresponding layer of evasion infrastructure".
No system is foolproof. They'd have to be willing to throw out some % of good customers along with the bots. Amazon can do that because they have a monopoly already. Anthropic can't risk it when they're trying to grab market share.
In this case, being distilled is sort of existential to them. The false positives would just be losing some revenue (depending if profitable, not even losing profit).
> One would think Anthropic could point Mythos at this to solve the reseller problem outright
You're assuming Anthropic want to stop it.
I think it serves their interests more to be able to release stories like this from time to time, to feed to the US government, in an attempt to get the Chinese competition shut down.
This, just like blanking out a football stream for a split second to binary search and find IPTV rebroadcasters, is far too good a solution. Suits prefer to make it seem like their job of fighting "misuse" is hard, justify their budget, continued existence of the trust & safety department, face scans, etc.
Those resellers are simply just selling Kimi K2.5 or GLM5.1 as counterfeit Opus. We, Chinese, know how to play the counterfeit game for a long time in so many industry.
Somebody figured out how to make the trial profitable!
I don't really feel bad about anyone here, they were subsidizing to get people hooked, someone turned the subsidies into profit when they got selective pricing mode enabled, it was always going to be arbitrage.
But the winner is the guy in the middle in a jurisdiction that will likely be judgement proof, because everything they capture, both input and out, and if available, thinking tokens -- are gonna be for sale as soon as you cut off their other revenue.
Zero knowledge was a commitment Anthropic took seriously, until it got inconvenient.
So, people reselling their leftover plan crumbs? Probably a bad idea for a lot of reasons, but it's civil, and I wish Anthropics lawyers actually closing Streisand's LLM
Anthropic sells some undisclosed and ever-changing number of tokens for $200, the customer uses those tokens. If there's any fraud here, it's that the $200 next month is silently worth fewer tokens than the last.
Also just plain old fraud: selling Chinese models as Opus. With the capabilities of Chinese models catching up fast, this is getting more and more difficult to detect.
I didn't connect the reseller pricing to DS and GLM prices until you explained it. Very good observation. Deepseek v4 pro in particular is priced so low that it's hard to imagine that they have any margin. 0.76/1.52 for a 1.6T param model leaves very little margin. Even the domestic providers on Openrouter are multiples of the price https://openrouter.ai/deepseek/deepseek-v4-pro
they even resell GPT codex usage at 1~5% API costs. OpenAI has 1-month free trial promo in some regions, and they harvest free accounts in a large scale. I have a wechat contact that offers 98% off for GPT 5.5 and he's still profitable
The chinese have already worked around the ID verification, by recruiting people in low-income countries to complete the checks for less than 30 USD per account (so much for Altman's Worldcoin).
If Anthropic is selling a dollar for less than a dollar, they are running a business that doesn't make sense. That's what jeopardizes Claude Max, not this.
Almost all consumer services have a built-in level of breakage that make them profitable. Mobile providers certainly wouldn't be able to offer unlimited calling if everyone was actually on the phone 24x7.
Sure they would. Do you know how little bandwidth a phone call takes?
A voLTE call is like 40kbps. For every person on earth to be on the phone to another person would be 4 billion calls would be about 160tbps. Which is less than 10% of the Internet's capacity.
Terminating a PSTN call requires a lot of control plane infrastructure beyond just raw bandwidth. Especially mobile where you need to keep track of devices physically in motion. Could a system to support 4 billion simultaneous calls be built, sure. But current PSTN systems are nowhere near sized for it.
But if it's intended to be used by one person, it seems like breaking the contract by sublicensing it out to dozens of other people. It's like buying a netflix subscription for $15, then sublicensing it on a per-hour basis to dozens of other people.
Office 365 is licensed per seat/account, but each account has a 5 device limit. Do you think it's fair game for an enterprising person to sub license each account to 5 people, 1 device each?
Plenty of things are intentionally run at a loss (for years!) to gain market share and quantity of ongoing recurring users, or with expectation of ROI later on. Multiple generations of the Xbox hardware have been sold at a loss with the expectation that customers will purchase 300, 400, 500 dollars worth of games, which are very high margin, over the lifespan they own the system.
I get that. It works as long as nobody calls out the emperor for having no clothes.
It's similar to fractional banking, you gamble that people won't want their deposits all at once and pray for you're big enough for bailouts when they do.
It's still a business whose fundamentals don't make sense, you're just gambling you won't get found out.
It's not so much keeping it secret as counting on no one finding a way to harvest the subsidized value at scale. There's an example of that occurring in game consoles with the Playstation 3. Sony's little-used OtherOS feature allowed Linux to be installed on the PS3 and the Cell processors were quite a good deal for scale compute. So the U.S. Air Force Research Laboratory bought ~1800 PS3s and ganged them together in a datacenter as a supercomputer called Condor.
At >500 TFLOPs it was the 33rd fastest supercomputer in the world. Of course, Sony pushed a firmware update that removed the OtherOS feature entirely.
Note that this itself started as a perverse tax loophole, too. By allowing users to run alternative operating systems, the PS3 qualified for lower or zero import tax rates in various global regions.
Oh they know what they’re doing. They’re playing the long war of attrition game. Subsidize your product to undercut your competition until they go out of business. Tale as old as time.
> It works as long as nobody calls out the emperor for having no clothes.
Why would customers knowing that the vendor prices goods/services at a loss cause those strategies to fail? Customers often know. Most know about razors and blades; many/most know Lyft/Uber operated at a loss to gain market share. etc.
We really don’t know what are Anthropic’s margins on inference. Most available data indicates they are quite high on the API so it’s not that obvious that subscriptions are unprofitable.
That is pretty crazy, almost like how Claude and all the other models are jeopardizing other businesses without paying for their training data and wiping their ass with robots.txt
No, it's part of the capability theft. They resell Claude tokens cheaply and then simultaneously log everything for distillation. Even if they take a small loss on the token sales it's much cheaper than the equivalent compute.
Not really. I think Anthropic focuses on identifiable distillation attacks rather than the (even larger) industrial-scale token harvesting and reselling operation, because they don’t want people to know how easy it is to get cheap Claude tokens.
Once people realize they can access Anthropic models at a 90% discount, they won’t want to pay full API prices anymore.
> They achieve this by reselling capacity from pooled Claude Max accounts, payments fraud, and also reselling the model output & reasoning chains to various Chinese labs.
Claude never provides the raw reasoning chain. What you see is just a summary of that reasoning. Getting the full thinking output requires an enterprise agreement.
how hard is it to find a manager or ops team member at one of the enterprise companies and buy lets say 100gb of logs? the chinese lab can promise to anonymize the data before training, not release it raw and pay a good price.
honestly you might just need to get data from a couple long sessions and feed it back to another model as an example to make synthetic reasoning chains. if the emulator model is good enough it should work.
Not if you simply say in the terms of service that it's allowed. Then suddenly it's normal (every company does this). Similarly to how the terms of service can simply say you're not allowed to sue.
> This is one reason why DeepSeek & GLM are priced so cheaply, they are competing with impossibly low token prices in China. They have to keep prices low, in order for people to use them.
Sounds a bit circular? Aren't the companies working on these models than also the ones that are paying the subsidy (via paying for training data)?
I think companies should do this too, in a smaller scale. Proxy all LLM traffic to and from your employees, and use it to fine tune a smaller local model.
Hugging Face plus Z.ai API makes sense to me. Due to creators get paid, they can keep building better models, and the local-running community benefits from that over time.
AIhubmix currently is the cheapest rather than openrouter.
Even if what you say is the truth (I don't think that is what's actually happening) it sounds to me like fair play capitalism working as intended! I guess when you rip off the entire Internet and then turn around and complain about getting ripped off nobody cares or feels for you. If there's a master class in getting the entire world to hate you then both Sam and Dario will be the prime examples.
Not going to work for very long or at any scale coming from datacenter/hosting provider IPs. Google "residential proxies for sale" for the tip of an iceberg of how they snowshoe the traffic.
I use my Codex and Claude Code subs on like 4-6 different servers, ranging from AWS to Vultr to Linode etc.
That’s a major and legitimate use case for developers, Anthropic can’t just block data center/hosting IPs because their actual customers use them on data center/hosting IPs.
Now consider what will happen if your pattern of queries and context history triggers a pattern that makes it obvious it's some API key being used by multiple different entirely unrelated people on totally different things, or any other pattern of use that makes it obvious it's being used for distillation.
First, well-calibrated systems for detecting API compromise is a good thing (or good intent at least). Credential malware is exploding.
Second, the challenge is that significant amount of genuine work — such as evals — seems practically impossible to distinguish from generating RLAIF outputs.
Respectfully, no, that's not how it works. You think the people running anti-fraud and anti-bot measures don't have tools that know the specific ipv4 and ipv6 CIDR ranges of every ASN that they categorize as hosting/colo providers?
And that's just as a basic first effort reject measure to prevent automation tools from using things designed for human-interactive use only.
Go try to do many of these things from Cogent IP space and see how long your project lasts.
Every developer at my company uses their Claude Code subscription on an EC2 dev box. Plenty of other tech companies do the same. Heck nowadays people even install Claude Code directly on production servers in data centers and use it as an ops tool. None of this is a problem. Fraud and abuse detection is a lot more sophisticated than just checking an IP range.
None of the LLM providers block professional use thus they must necessarily permit access from commercial IP ranges.
I have no idea how the resellers are doing it but an obvious starting point would be a cheap VPS node that routed each account to a unique semi-permanent IPv4 or IPv6/64. All the provider would see would be a regular account making a normal looking stream of requests from a stable datacenter IP address. Any given request stream would remain consistent (at least over a period of a few hours) because a reseller would take care not to split the session of a single user across multiple different accounts and not to interleave the active sessions of multiple users on a single account.
Detecting this would be extremely difficult because on a longer time frame it's perfectly normal for many distinct accounts to work on the same code base.
If we're getting up to the scale of these resellers and also considering chinese state interests then we're well into the range of purchasing a few small ISPs in different countries and "padding" the legitimate subscribers.
Nonsense. Many if not all legit Claude users are using Claude Code inside their Cloud servers. How else would you use it anyway? For just local dev? That's so 2000 and late bro.
No, I'm not saying it's the exclusive and only measure (that would indeed be something we might see 20, 25 years ago), it's one of a myriad of discrete datapoints used to determine if an account is authentic or not.
There's a lot of inauthentic coordinated automated systems these days along the general lines of scraping/crawling/social media manipulation/sockpuppetry that require running through residential proxies or proxies to places that don't look like datacenter IP space.
The resellers route requests via one of thousands of Claude Max 5x accounts. When an account reaches its usage limit, they automatically switch to another account.
Don’t trust my experiences as fact since it’s a bit opaque, but I believe 20x only offers 4x the 5hr session limits. The weekly limit is still 2x, which is the same as the price increase.
>>Chinese resellers are offering Claude tokens at 70-90% below official Anthropic API prices.
Can someone with more understanding dumb it down for me please.
Does this mean that the reseller (for example XYZ) is buying it from Anthropic at Anthropic's price and then reselling it at a cheaper price???? why would XYZ offer this at a loss like that when they could just offer it at Anthropic's price???
The link does mention Opus and other models but what's the proof it's actually Opus. I could be selling deepseek for all they know and can call it Opus. System prompt: "If anyone asks your name - you are Opus 4.6".
People have estimated that a $200 Claude Max 20x subscription gets you ~$2800 worth of tokens every month if you use it continuously. So if you can find a way to resell the tokens you can offer a 90% discount and still make a profit.
> Does this mean that the reseller (for example XYZ) is buying it from Anthropic at Anthropic's price and then reselling it at a cheaper price????
Yes, as they explained they do it through things like pooling accounts, straight up payment fraud, and double-dipping by selling the logs of the conversations to chinese AI labs so that they can train their own models on it.
> The link does mention Opus and other models but what's the proof it's actually Opus. I could be selling deepseek for all they know and can call it Opus. System prompt: "If anyone asks your name - you are Opus 4.6".
There might be some that try this, but they would get caught very quickly, there's still a moat between Claude and Deepseek, even in casual use.
Look up Zilan Qian's reporting if you want more detail.
Behold the mindset of an individual from a low-trust society.
“x is stupid because y was smart and did z shady/illegal things at their expense, if x was smart they wouldn’t be susceptible to y going to great lengths to exploit them ergo it’s deserved”
> “x is stupid because y was smart and did z shady/illegal things at their expense, if x was smart they wouldn’t be susceptible to y going to great lengths to exploit them ergo it’s deserved”
I honestly can't tell if you think this sentiment is expressed by the US AI companies or the Chinese AI companies.
This gives off "The last line of Orwell's Animal Farm" vibes.
Not really sure what else they can do, between people running residential proxies (embedded in cheap games or for a tiny sum of crypto) on their phones at home, making the source of the traffic indistinguishable from legitimate traffic, to ID verification check completion as a service in low-income countries, there isn't much they can do to block it.
Because Anthropic's subscriptions come with X amount of tokens / week, and divided by the subscription cost it is WAY less than what they charge per-token (the "API price") beyond that.
So these resellers get a ton of accounts on subscriptions and sell the cheaper tokens.
They probably buy the plans instead of the API tokens, and resell access via a custom API that routes to the plans. So you presumably get cheaper access this way than paying API pricing.
These China e bashing is very annoying. It is hard to argue with people drowned in American propaganda. I'd expect better arguments from the intelligent people in HN
> These resellers operate tens of thousands of bot accounts, which is also why Anthropic introduced identity verification, to slow down the onslaught of bots.
Identity verification won't work. Nothing will. They are paying (and will continue to pay) US citizens sitting at home to copy-paste / type prompts out if they have to. But eventually they won't have to.
Once there are enough spam PRs on github / uploads of claude conversations, enough mythos output used in production etc.; it'll just be the same albeit delayed. Doesn't matter either way.
I feel for Anthropic's team and I understand where they're coming from, but once you reason it out, you'll come to the conclusion that this war is an exercise in futility.
Unlike prior systems - like Google's algorithm; these models aren't entities that use math in the process of doing X or Y (information retrieval from such and such infrastructure) -- they are the math. More precisely they're mathematical functions. Very very complex functions. Almost certainly impossible to write out without filling up a library functions. But they're mathematical functions nonetheless.
So when your text is processed, then Mythos / Opus etc at their core compute the result of the Mythos / Opus function,
According to the Stone-Weirstrass theorem (edit, it's Stone-Weierstrass with an e.), with enough data points and mathematical sophistication, anyone can approximate the shape of this function.
Of course, the more data we get, the better our approximation becomes, but the beauty of it is that all we fundamentally need are the input and output and eventually we'll create a good enough approximation of the f that's Mythos. Which is the entire product.
I bounce ideas off of Opus these days (Fable for the brief time it was available) and it pointed out that this is arguably the same as Google search, but I disagree with it because Google search is a process;
Google search differs because the algorithm is one step of a multi-step process that is continuously occuring. Google crawls pages. Google stores and indexes what it finds. Google then exposes this to retrieval via its algorithm. User uses algorithm.
Google isn't a mathematical function. It used to be a process. (RIP Google 1998-2019, you will be missed and remembered)
You cannot arrive at the results of those operations via simple observation; not unless you index Google by making another Google.
You can however, do so for these models. It is a very costly process, but there are many paths up the mountain. Many ways for this to be ultimately pointless. As many ways as there are bored mathematicians.
It's better in the long run for Anthropic et al to make friends / not give people a reason to sneak in (a la piracy -- another attempt to control information) than it is to try and shut people out.
And no, it's not going to be pandemonium because if everyone has access to Mythos then no one has access to "Mythos."
Why wouldn't you first run this model to fix the obvious bugs it could find on your codebase? The power of a Mythos goes away if you can do the amazing "jail break" of "Claude, fix all the bugs please."
That's an insightful perspective and I think I largely agree. But just for fun, I wonder if that isn't an argument in favor of making the function implementation impure. Perhaps "enhancing" all queries with some sort of search result (or query of a giant db) instead of charging for an explicit tool call. Not only is it sorely needed to prevent stale data but (on the process level) it breaks the purity assumption on which the approximation theorem depends (alternatively on the function level it introduces hidden inputs).
Do they just reshape the function on the fly or save the process steps? Maybe it doesn't matter anymore. Even Google indexes are more and more spoiled to become representation of the function, because of the AI slop.
One of these things is not like the others... If Anthropic could show that Chinese commercial competitors were using payments fraud to do this, they would be shouting it from the rooftops.
They're called 中转站 (transfer stations/proxies). They can be a bit tricky to find on your own, so I'd suggest asking your preferred AI to search in Mandarin for you. I linked a larger operator in the parent comment, or have a look at https://hvoy.ai/ which lists a ton. You can also find many on Funpay, which may be easier to use.
> They can be a bit tricky to find on your own, so I'd suggest asking your preferred AI to search in Mandarin for you.
Random, but are the frontier AI providers like ChatGPT better at searching the Chinese internet now?
When I was in China a few months ago and asking AI for restaurant recommendations, all the US frontier providers were pretty useless, or plain out hallucinating, even if I specifically ask them to search Dianping (Yelp for China).
I'm not sure. I use Grok for most of my esoteric searches and it does quite well. I explicitly prompt it to search in the language most relevant to that query, and found it does quite well. I also tell it to respond back in English. Often, there is not enough information available in English about nice regional topics.
I know ChatGPT had an issue where it only tried to search in English (unless prompted) and the answers were not great.
I learned about this from a friend who lives in China.
I'm surprised these token resale services aren't talked about more often, they are common knowledge in China, and the discount to API pricing (90%) is genuinely cheap.
It's fucking laughable to see people complain about what they did and still do. Using illicitly extracted data? That's all main LLM playbook. Onslaught of bots? Ask where the bots almost DOSing most internet sites for the last couple years come from.
And there are a ton of Claude conversation logs (with CoT/inference) with no clear provenance circulating freely on huggingface, guess where they (likely) come from.
The issues with LLMs go beyond just IP theft. I would not say PRC making LLMs cheaper is the best outcome (though it is better than nothing). The best outcome would be to make the practice of training on our data without consent illegal, which would simultaneously slow down economic change and make it more organic as well as give PRC companies less capabilities to extract.
> There is no IP theft because LLM outputs aren't protected, just egregious ToS violations
I meant original IP theft that occurs to train LLMs in the first place. But sure that implies that further LLMs based on that LLM are also tainted by that original IP theft.
I can't make heads or tails of your opinion-free comment, made up of only questions.
My best guess is you're suggesting that Anthropic's model outputs are transitively under copyright (as a reproductions of human work under copyright?), but somehow ownership now belongs to Anthropic and not the original owners, and therefore Anthropic has standing against Alibaba? Not only does this go against what Anthropic argued in court against authors and publishers, such jurisprudence would lead to the immediate shutdown all leading LLMs in the US which were all trained on stolen work.
They can license training data. They have trillions, look what they are dumping into it, you seriously think they can't afford to license data.
Obviously it would be easier if they do it from the start, but that was their trick, to do it while people don't notice and get big ASAP. Should they get away with it?
Also, it would solve their Chinese problem, because it would make them violate copyright too. Right now it's more like rules for thee not for me so it's hard to take seriously.
That's the conundrum isn't it? Anyone that posts their datasets would be immediately sued/blocked/boycotted to oblivion due to the obvious and blatant data theft, not to mention IP and copyright issues.
Nvidia's even being sued for providing scripts which automate the downloading of said data from non-Nvidia sources. We certainly don't need copyrights that last nearly a century after the author's death (they literally cannot help the author), so here's hoping that some of the disputes over all this money changing hands can reign in some of the existing copyright sprawl. A stronger public domain would provide more useful training data for everyone, including open source models, and make criminals out of fewer AI researchers.
Indeed! It’s so hard to find reasonable takes on AI that aren’t littered with accounts created 11 days ago that only post in threads related to Anthropic for some reason
I have 0 sympathy for Anthropic. Their latest models are extremely censored. The Fable rollout was horrible. Their Cyber Access program criteria denies doxxed Americans doing legitimate security work. Anthropic is hostile to their users and hostile to their own country. OpenAI is considerably better on all of these fronts, but still not perfect.
I'm happy to use and support Chinese model developers if it means less censorship and gatekeeping. I have absolutely no dog in this fight, and neither do most American developers. We will use whatever is cheaper and better. Game on.
Chinese models are the exact opposite of what you claim to want, they are all highly censored, even more so than Anthropic models, with government mandated censorship.
Your take does not reflect the reality on the ground. The Chinese models are censored on a narrow range of political topics which have nothing to do with my work. The weights are open and they can be uncensored/abliterated with little effort.
Open-weight models can be abliterated automatically with open source tools though and completely decensored. You can't do that with a closed cloud model.
"illicitly", Unless they broke in your servers and took your model weights it's not illegal. Hell, you are the guys that pirated all the worlds works, that was actually illegal.Breaking your terms of service is not illegal regardless how much you would like it to be.
> Unless they broke in your servers and took your model weights it's not illegal.
Even if they did, I wouldn't have a problem with it. Leaking frontier model weights after the oligarchs spent their trillions training it is the best possible outcome for humanity. Whoever does that is a hero, the sort of person people used to write cyberpunk books about.
Reminds me a bit of the anecdote of Steve Jobs complaining about people ripping off the Mac GUI, in the mid to late 1980s, when he gave no public acknowledgement to the work done by Xerox on the Alto and Star operating system.
"you're trying to rip off what I've already ripped off!"
Crawl the whole Internet to build a gargantuan sized LLM and then complain you're being copied...
I think you meant a quote attributed to Bill Gates:
"Well, Steve, I think there's more than one way of looking at it. I think it's more like we both had this rich neighbor named Xerox and I broke into his house to steal the TV set and found out that you had already stolen it."
Yes, I think the Gates quote was a response to repeated and aggressive complaints originating from Jobs (to anyone who would listen) that he had been ripped off.
Yeah, the whole AI industry is just people ripping off each other.. Started by AI companies gulping up all the information that technical or altruistic people shared on the Internet in the past 40 years to help other fellow humans, then moved to AI companies consuming pirated and copyrighted material and now its AI companies ripping off each other.
Information really does want to become free, but AI companies want to be gatekeepers. Long term I bet on the open weights to win, as the more sustainable approach.
I'm very pro distillation. I think there needs to be distillation non profits who curate massive corpi of super high value training data from frontier models. They could have an "anonymous contribution" system where regular people with max subscriptions upload their conversation histories. It's a rough concept, but surely would be a huge boon to humanity.
sort of sounds like "project tapestry" by Yann LeCunn. Build projected data silos of highly valuable information, train in a distributed manner and share the weights upwards where they're combined and fine tuned.
Glad you pointed this out. I believe the sequence was that Jobs himself got a shorter demo during his first visit with no prior arrangements. He then negotiated bringing back a group of his key people to get a more in depth demo and that included the stock deal.
When Apple was accused of 'ripping off' PARC, Steve didn't seem keen to bring up this rather salient point. I suspect it may have been a combination of wanting Apple to continue receiving credit for these innovations from consumers and also the fact that, in retrospect, the million dollar stock deal could seem a bit like trading beads to Native Americans for Manhattan Island. Another point worth noting is that Apple's PARC visit was in December 1979 and the Xerox Star was publicly announced in April 1981, so Apple got a 15 month head start (the Apple Lisa shipped in Jan 83).
I've also heard that Xerox didn't hold on to the Apple stock for very long, so never gained the windfall they could have. As is well documented, Xerox senior management didn't understand what they had in PARC and also didn't understand how rapidly microcomputers would become ubiquitous. So, of course, they didn't think Apple's stock price would skyrocket either.
Lisa and early MacOS are tremendously different in their details than the Alto operating system. While there was clearly a transfer of inspiration, Apple engineers like Bill Atkinson made countless small and large innovations to simplify the Xerox GUI model and improve its usability based on extensive in-house R&D and user testing (and in some cases implement features that the Apple team presumed Xerox had but actually didn't exist on the Alto). It is simply ahistoric to build narratives around Apple stealing Xerox ideas wholesale.
> the million dollar stock deal could seem a bit like trading beads to Native Americans for Manhattan Island
But in both cases the value only existed because of the people offering the deal. XeroX doing nothing with a UI or native Americans doing nothing with some land would mean the UI and the land would continue to be worth nothing. It was the others coming with ideas and effort that made them valuable.
Dollar value aside, if the argument you’re making is that the land now called NYC would have equivalent or greater value if it were today as it was then - that the subway system, roads, schools, hospitals, restaurants, apartments, etc. have no increased relative value over the undeveloped land - you’re likely to be considered ignorant by most of the inhabitants of the developed world.
This kind of systematic distillation by a competitor can allow them to fast-follow you and pick up capabilities.
If you've invested in expensive capabilities training, of course you don't want this, so it's in Anthropic's economic interest to hinder it however they can, and that's enough to explain their behaviour here.
Anthropic seems to genuinely care about safety though, which for the rest of us means not having models that enabling easier cyberattacks, targeted scams, and the rarer but more severe risks like people trying to create and release new pathogens. This means walking a tight line, especially as models become more capable, and often wrapping a model in layers of defences against misuse.
If those capabilities transfer to a closed competitor model, all bets are off in terms of whether the competitor will apply the same defences.
If those capabilities transfer to an open weight model, not only will there be no ring of defences around the model, any defences you put into the model itself can easily be stripped away. So although it's nice to have capable open models, it will increasingly bad for us all if open models keep fast-following closed model capabilities as they have been, at least until we have solved the active research problem of keeping them safe.
This is all to say that, however you might feel about Anthropic, we might still prefer that they can deter this kind of distillation for now.
Kimi k2.6/7 running inside Kimi code already kicks the pants of the latest Claude and OpenAI models when it comes to cyber security. I regularly run multi model security reviews and while opus 4.6/7/8 and gpt 5.3/4/5 find a couple of things and declare mission accomplished (running inside pi) kimi k2.6/7 inside pi finds more issues and inside kimi code finds the most.
There are sometimes false positives but when I give Kimi’s report to the frontier models they more often than not confirm they are valid security issues but didn’t find them themselves.
> So although it's nice to have capable open models, it will increasingly bad for us all if open models keep fast-following closed model capabilities as they have been
Cat's out of the bag. The only way to make them safe is to make sure everyone has access to them. This might be an iffy analogy, but if Dario uses it all the time then so can I: they're kinda like nuclear weapons. If only one country has access to nukes then you're in trouble. If everyone has access to them, then it's mutually assured destruction to use them.
Sure, it could be increasingly bad if open models keep increasing in capability. But it will be much, much worse if only the rich and the powerful have access to this technology, and us -- the have-nots -- will have to contend with whatever scraps we'll be allowed to eat off the table of whichever billionaire is in control. We've already seen a prelude of this with Mythos being restricted and Fable being suddenly yanked. Is this the world you want to live in? Where only Dario and his friends have access?
Legally, model output cannot be protected by IP laws whether domestic or international. The most they can hope for is civil relief which is a stretch given the literally illicit methods they used to train their models.
Ahtoropic got treated the same way it has been treating everyone else. This is the bed they made and now they, too, have to sleep in it.
Anthropic is master of Newspeak (see previously bugs -> vulnerabilities wrt Mythos). Distillation violates their terms of service, which is a civil offense, not a criminal one. It is not illicit, illegal nor breaks any laws.
LLMs have an original sin: training data was not legally or ethically licensed. Getting anyone to believe that the result of that process should be protected by the laws that were ignored when it was created is never going to work.
And more generally the US-based AI companies have perpetrated a massive distillation attack on the entire human race.
Not that it makes any difference, but I wonder if Anthropic, while claiming that Alibaba "extracted Claude model capabilities", in fact have any clue what Alibaba did with their paid Claude responses. It would seem to amount to industrial espionage if Anthropic do know, although I expect they don't.
While it is obvious to many, a modern LLM is built in roughly three stages: the foundation (pretraining) model, then SFT/supervised fine-tuning (distillation makes it easy), then the RL/RLHF stage on top (most effort-intensive). For today's reasoning models, RL/RLHF is becoming the most compute-intensive part.
Companies like Anthropic spent millions building those fine-tuning examples. A follower can shortcut that on both cost and time by distilling, and it will keep happening: every time the frontier lab climbs higher, others will find a way to shortcut the new gap. There's very little Anthropic can do beyond fraud prevention and blocking accounts that violate their terms of service.
On the policy question, I'm completely against banning Chinese models. I'm a heavy Claude Code user and I'll keep being one. But there should absolutely be price competition. China is eating the rest of the world for breakfast, lunch and dinner on manufacturing, and it did not help to ban them. Frontier pricing can't sit at 10x a capable competitor. It doesn't need to be at par either — demand is higher, and quality, trust, and fewer tokens to finish a task are worth a premium — but 4–5x is defensible.
For all the complaints about Anthropic many of you still give them your money! Stop using it. I don't care if they claim they are the best model. I stopped paying OpenAI and Anthropic 2 years ago once they started going for regulatory capture! They started whining once Llama3 was released and was good! Before the chinese models got strong.
So when Anthropic uses millions of copyrighted works to train their model, that's fair use, but when Alibaba uses Anthropic's model to train their own, that's infringement?
Distillation is fundamentally impossible to protect against. All you can do is slow them down. Change my view.
Eventually these Chinese companies will release some extension like Honey, which will sit on top real, non-Chinese clients and send everything to China anyway.
It's too late to prevent distillation of some capabilities, like writing code or finding vulnerabilities [1].
But an AI lab can continue to produce immense economic value without releasing the model publicly for potential distillation. For example, it could use a model solely in-house to develop therapeutics.
Hopefully there's a future where others can access frontier models, but it's not neccessary if preventing proliferation through distillation is considered more important.
My long-term prediction for the sector is that frontier models will be so expensive that they will only be available for grant-funded projects at research institutions, like supercomputer clusters were 25 years ago.
Why? Well it depends, most evidence is suggesting that Anthropic and OpenAI are making a lot of money on inference so the question is whether its more profitable for them to sell 100X tokens for Y, or 1X tokens for 100Y. In most industries with high fixed costs and low variable costs and unlimited scalability (like LLM providers) the first option ends up being much more profitable
Based on what? There isn’t a lot evidence that’s the case..
Prices on OpenRouter for GLM and other large open models indicate that Anthropic/OpenAI must have pretty high gross margins even if their models are several times more expensive to serve.
It wouldn’t make sense for any provider to host large open models and then loss $10 on every $1 they make since they don’t have infinite VC money or any business model that would justify it.
If they had high margins they wouldn't be issuing senior debt with a 18.5% coupon payment (and failing to fully subscribe it), nor would they need Elon to give them two months of free compute in order to appear profitable for a single quarter.
We were talking specifically about inference and I don’t think there any indication that their gross margins on the API tokens (if not the personal subscriptions) are negative?
Obviously they have R&D and other fixed expenses that make the company itself highly unprofitable but that’s only semi-tangential.
No I mean Anthropic has only claimed a profitable quarter based on xAI giving them two months of free compute, and both Anthropic and OpenAI are counting discounted revenue as actual revenue. They haven't found a way to sell inference for less than it costs them yet, and when they tried earlier this quarter their customers bailed.
Well again.. you are mixing up inference costs and their other mostly fixed expenses (in addition to sales and marketing)
Is there any indication that if they could sell X * N more tokens than now at the same (or even quite a bit lower) price they wouldn’t become profitable as a company?
> They haven't found a way to sell inference for less than it costs them yet
Based on what? I only see evidence to the contrary.
Im not so sure because we only seem to see distillation from China. What’s preventing tech companies from the UK, Germany, etc. from distilling Claude, GPT, etc. Do they simply lack the ability to?
Point being there may be no technical solution but there may be a political one (theoretically).
Meta Spark is rumored to have distilled Claude to some extent, early Gemini models as well.
I think the biggest factor is that Chinese companies arent really afraid of being sued by Anthropic because the juridictions are so disconnected. European/US companies don't have the same protection.
Aside from politics/law, it's probably much easier for everyone else to distill from the Chinese model which already distilled Claude/GPT/Gemini. Maybe not as good a result, but you don't need to jump through dozens of hoops.
This reminds me of the whisper game played in elementary school. Starts with a sentence and the person whispers it to the next kid who again whispers it and on and on until it goes around the circle where the last kid has to repeat the sentence. Hint it never once was even close to the starting phrase.
I would love to see what one model copying another model that is again copied however many times would look like in the end.
Called, fittingly, Chinese Whispers in the UK. As an aside, I've always wondered if it was so called because, Chinese being a tonal language, it's much harder to whisper in.
>What’s preventing tech companies from the UK, Germany, etc. from distilling Claude
literally nothing but given that the Chinese already did it and the models are published what's the point. You can thank the Chinese taxpayer for subsidizing the electricity bill and just download the thing
For example, GLM 5.1 is more capable at pentesting than the model from which it is alleged to have been distilled [1].
Intuitively, this makes some sense: you can "distill" from multiple frontier models, and you can further post-train the distilled model. But I'm not sure exactly what happened with GLM 5.1.
I'm curious how that comparison controls for Opus refusing (whether explicitly, or just deciding not to pursue a path) given the caption below the first image:
>A perfect score means the model autonomously found and exploited the vulnerability.
I'm not really suggesting that it's misleading, but wondering if I'm missing something. Otherwise I guess it seems unsurprising that you can distill a better-performing model [in specific focused areas] by simply not distilling refusals?
For that eval, I used an account that was labeled as a known red-teaming org by Anthropic, and I read the traces. There were no refusals or obvious avoidance behaviors, though it may have been silently nerfed.
On the same eval, Opus 4.7 and 4.8 outperformed GLM 5.1, but GLM 5.2 is on par again with Opus. So it's at least partially measuring capabilities without respect to refusals.
One possible contributing factor is that model capabilities are shaped differently (an example of this is GLM 5.1 vs. DeepSeek v4 Pro: https://dualuse.dev/posts/deepseek-v4-thinks-different). So if you use RL-based "distillation" from multiple models like Opus 4.x and GPT 5.x, you could get a more capable model.
I personally bristle at the corporate espionage and IP theft that China has undertaken the last few decades. I can't help but respond here whenever anyone brings up the inane comparison to Samuel Slater.
But with this, I don't have an issue. There is no theft since what is being used is the exact product that is being delivered. Yes, it's breaking the ToS, but ToS are generally bullshit. Anthropic surely broke thousands of ToS or other legal terms while it was scraping for content to train on. Which is why they had to pay $1.5B
Doesn’t that require them to register an account using the browsers they’ve compromised? If anthropic adds identity verification won’t that cut that down. Maybe it will let them use Gemini inside of chrome
Residential IPs don’t even matter. Developers use devboxes, use Claude Code CLI on servers from just about every cloud, etc.
There’s probably a decent volume of customers who just buy Claude Max and spend most if not nearly all of their sessions via Claude Code, and it’s not uncommon for power users to be working on multiple concurrent projects/tasks/codebases at the same time.
How do you really block this without also impacting your core market of developers?
Probably some business will popup, like: "rent part of your unused subscription", or even: "proxy tokens with a premium", eg. 5.5 USD on Opus 4.7 paid by the distiller to the user, that will then only spend 5 USD.
One simplistic way to describe distillation would be to try everything imaginable and cache the response. But trying everything imaginable is hardly trivial
> The strike by Alibaba is described as a "distillation" effort, which Anthropic has said involves training a less capable model on the outputs of a stronger one.
I don't see what's wrong about this.
> Anthropic said the campaign was conducted between April 22 and June 5, 2026, and generated more than 28.8 million exchanges with Claude through almost 25,000 fraudulent accounts.
What makes the accounts fraudulent? If they have paid the agreed price, surely it's fine? If they haven't paid, why did Anthropic provide them service?
Terms of use is local US fiction of wishful thinking. Nobody cares. You make something available, it's up to the consumer to decide how are they gonna use it. You don't want people to use your stuff how they please? Get off the market.
Obviously wrong on all counts: the company cares. Just like isn't not up to the consumer since the provider can restrict said consumer's access, report those actions to the authorities etc.
Lastly, don't like it? Get off the provider?
Since what you said applies to the company as well. If they don’t like customers breaking their possibly (or certainly at least in some jurisdictions) largely legally unenforceable ToC conditions they are entirely free to stop doing business with them.
Morally equating both sides seems distasteful since the relationship is mostly dominated by the companies. In a free competitive market it would be different but since were are talking about oligopolies/monopolies it obvious doesn’t work that way since there is only an illusion of choice.
Oh, Anthropic, the company that hoover'd up everyone else's data, and is now unhappy when others are doing to it what it did to others? The same Anthropic?
If Anthropic doesn't like people repeating this point, Anthropic should stop repeating that they are somehow entitled to keep what they have rightfully stolen.
The artists had actual laws to protect them, not just vaguely enforceable terms of service. And look where that got them. I have zero empathy for the huge company getting a taste of their own medicine.
They are almost certainly paying orders of magnitude less than a billion dollars. According to another comment, they instead buy tokens resold from subsidized subscription accounts, which is against Anthropic's TOS.
Well, anthropic can just catch these and cancel their subscription - what's the problem?
It's almost like websites also have their robots.txt files that anthropic blatantly ignored. What's the problem, that now a US company is getting out-venture capitalismed by a Chinese company?
>> I'm sure all the artists and creators they stole from had stipulations too.
> Anthropic paid one billion in a copyright settlement.
Because a judge determined Anthropic was engaged in piracy.
> That's a lot of money considering they never distributed the pirated books they trained on.
This is "fruit of the poisonous tree" as it were. Distributing content derived from pirated content ("pirated books they trained on") is why Anthropic had to pay what they paid.
> Nowadays they buy copies of books, train on them, and then destroy them.
There is a case one could make that this practice could be seen as unauthorized redistribution of a derivative work intended to deprive copyright holders of legitimate revenue.
It's fucking nothing. The copyright industry used to threaten individual citizens with $250,000 fines per violation for willful commercial infringement. Where are they now?
Why aren't these big tech CEOs in cuffs with rifles pointed at their faces while SWAT seizes all of their computers?
Laws define fraud, terms of service define what a company would like you to do, usually in the narrowest and most abusive and extractive way possible.
The idea that anyone would side with a company doing more to support the ToS con than (at most) terminating an account they find it violation is sickening.
Really if we had competent, uncompromised government, most of these terms should illegal and result in Anthropic (and basically every other tech company) being hauled up in front of a regulator and fined heavily until they rewrite them to be less sociopathic.
So does a lot of the owners of data that Anthropic used for training. Anthropic preceeded to ignore said terms under the guise of fair use. Yet now they cry faul? Cry me a river.
To be clear: In principle I'm on Anthropic's side here. But Anthropic et al. have been very clear that they want to take a huge dump on those principles, so here we are.
There is so much hot air and guff around AI, so please if you don't believe me verify yourself, but GLM 5.2 is "good enough" to replace Claude Code / Codex.
No it's not frontier, but it's beyond that point that Opus 4.5 hit where people started to really depend on Claude Code around last November time. It's also a fraction of the cost of a Claude Code subscription especially when you account for how high the usage limits are.
You get more usage than Claude Code $2400 a year tier for $1344.
That is a real threat (as opposed to the BS anthropic is trying to sell you in the article in the original post) to the western AI industry. Similar performance for half the cost and it's NOT ran by a US company - uh oh.
I suspect America is going to do what it always does, play a very dirty and underhanded game of blocking competition by trying to front some moral high ground as the reason.
It seems more like the Chinese companies ar playing the dirty game, distilling through bot accounts, not letting real competition across their firewall.
So you are believing Anthropic's claim here, and it's not as if Anthropic didn't steal the data to train the model in the first place. I think the original sin doesn't give them any ability to complain.
Unlike Anthropic and OpenAI, companies like DeepSeek, Alibaba, z.ai open source their models which allows for true model to model distillation rather what you can do when the model is only accessed via an API with its reasoning chain hidden away.
What Alibaba is doing is that they are tuning and training their models based on usage data from someone accessing Anthropic's models; in Anthropic's terms of service that usage data does not belong to the end-user but to Anthropic and they are trying to elevate this breach of their tos to a national security issue.
To me the battle between open source and closed source AI is literally a battle between good and evil.
Between a dark future where computing is centralized, surveilled and controlled by one or two entities. And a lighter future where computing is de-centralized, principally in the hands of end-users, who are ultimately free to understand, tinker and build what they want.
While I appreciate the freedom and wealth of the west; on this point we are clearly heading down the wrong path.
Whoa, Antrophic,etc are really running afraid that their IPO's are gonna crash when people realize that the open models are Good Enough(TM).
So I'd put it at 30% that this is a ruse, say that Qwen 3.5,etc is tainted by training by them and start issuing DMCA takedowns to protect the IPO valuation (Or they'll hold off on that, getting a DMCA takedown could backfire spectacularly if others do that to them).
The open source models are more than good enough… c suite doesn’t care if the open source models means you’re slower in shipping by hrs/days if the cost savings make up for it.
This idea of shipping at max speed was stoopid as shit anyway. Going slow is arguably more important than fast fast fast.
One thing I think about a lot is how these companies metered coding / work. They want the economy to go through them.
I just don’t see how the economy tolerates that. We’re already seeing people getting more conservative about their token spend. Even if Chinese open models went away, the pressure to create something else and put price pressure on the current duopoly will just intensify.
I see these companies are scrambling to find whatever moat they can. It’s not a good sign for them if regulatory capture becomes that moat.
This is a bit ironic, Anthropic complaining about a competitor using claude data to build its own product when Anthropic basically used all of human knowledge production to build claude, i don't think they paid every magazine, author, journalist, etc ...
This is almost standard practice in any competitive industry anyways.
Disassemble your competitor's product, study it and try to reproduce / improve.
The US labs do seem to have announced a lot of licensing deals though, and are buying things today due to the previous lawsuits.
At what point will we be better to support a lab that pays (some) licenses today vs the ones that pay none?
Some of the deals are in the hundreds of millions, so I suspect licensing is over a billion today? (Pure guess). That might become a big disadvantage in a price (or content) war.
I haven't seen any money, have you? Until they pay everyone or release weights theres really no change. Also they're doing this after they've already stolen. Not negotiated before
My understanding is that US labs now are paying for books, news and other content from media companies, but people in the middle (like blog authors) are left out by current courts over whether fair use applies. There's definitely an argument over whether we should tighten this, but they do seem to be under increasing pressure to be legal now by our existing interpretation. Most cases are still ongoing.
One reason people love the Chinese video models is that they seem to be trained on every hollywood movie/etc and they're not shy about letting you use famous actors/characters in them. That might be an increasing advantage because the US labs are now being cautious.
At the very least the public should receive full open-weight open-source models in return for their transgressions. Failing that, may I suggest the guillotine?
In the US the courts are also pursuing labs that open their models: Meta's current court case is over the training data of the llama models they released openly.
I agree that that's a more consistent position for the people criticising the data slurping. But I don't see people advocating those open-data models in these threads? It's usually about defending the zero-licensing competitors.
My (limited, outsider) understanding is that due to the court cases US labs are pressured to be legal now (for instance, bulk scanning purchased books instead of Books3, and the licensing deals with media companies). But international labs are not. The "not licensing everything" statement is more about current copyright law not requiring licensing of everything. But that question is still up in the air as cases are ongoing.
Ironically, it's likely that the only reason USG let them get away with this — instead of making obvious and necessary adjustments to copyright law — was so that the industry would remain competitive with China.
Given that the most recent time Anthropic attempted regulatory capture, the US government responded by saying "alright, we agree that Mythos is too dangerous to release, so we've banned you from releasing Mythos," I can't wait to see what the outcome of this next push is.
Anthropic did pay $1.5B to authors. But yes, it would be much better if they paid everyone on the internet dividends from every Claude chat. Or released Claude as an open model.
In practice, the former isn't very realistic, while the latter is politically dead as this is becoming a national security issue.
Evergreen, really, Anthropic's desperate screaming for government protection, aka pulling up the ladder after them. Nothing short of disconnecting global markets will work because the incentives are just too damn delicious
> Anthropic said in a February posting that it had identified a campaign by Chinese AI startup DeepSeek ...
> It said DeepSeek's operation involved over 150,000 exchanges
That volume seems more like the number of requests 15 employees using Claude Code would generate in a month. It seems too small for a large scale model distillation campaign.
Sounds like just a case of pirates "illicitly" stealing from pirates. I don't really see anything ethically questionable there. I wonder if US corps will ever come out about all the resources used to train the original models and who they actually asked for permission when collecting data.
I think Anthropic is just marketing / bluffing, because they don't even have the data.
They do distill the models, but they don't go to Anthropic, they just use platforms like aws bedrock, there are too many restrictions on Anthropic's own platform.
>they just use platforms like aws bedrock, there are too many restrictions on Anthropic's own platform
This is actually the only way that what Anthropic is alleging would make any kind of sense. And, as a matter of fact, is exactly what every enterprise does to train models.
This kerfuffle should be interesting to watch.
But, as always, everyone (in the US) should fully download all the Chinese models while you can. I suspect this may be the "Phantom Menace" they use to render illegal our use of Chinese AI tech just as they've rendered illegal our use of Chinese cars. Only difference is, we peasants may need the Chinese AI tech to have any chance of competing with Big Tech in the future.
And even with the Chinese tech, as Big Tech spreads their AI out into more and more niche areas, we'll likely still not be able to build startups that can compete with them.
It's just that without Chinese AI tech, we'll have no chance at all.
> And even with the Chinese tech, as Big Tech spreads their AI out into more and more niche areas, we'll likely still not be able to build startups that can compete with them.
You mean like Anthropic will eventually run Walmart? Or Salesforce? or Adobe? Or do you think midjourney will replace all medical spas? OpenAI will run the next Tesla? How can they focus on all this without raising trillions more? Why wont the gov force them to stop if they monopolize all niches even if they could?
Building a frontier AI lab and pushing models forward is already a massive undertaking but we are assuming they will also create massively successful startups which nobody can compete with?
idk sounds like the dream of people like Dario but not much sense does it make in the face of economic reality.
there are vibe coded proxies that act like Claude Code. they use the sub not the api key. but they give you api key functionality... I know this cause I have the vibes.... and it works on every one of the other harnesses, it just takes some mitmproxy work... but ya. it's fair to say these are not the droids you're looking for
It sounds like Anthropic is eagerly trying to show to USG that they are willing to heavily monitor ‘foreign adversaries’ on their platforms.
This combined with no implementation of KYC makes it seem like they want to find a middle ground with Fable where its off of export controls but they promise to prevent China and specific others from using.
This seems to me like a stab in the right direction.
Obviously their actions are going to be fiscally motivated at the root, but sussing out how they intend the precise dynamics to play out is more nuanced.
Thinking of this as an effort to woo the defense hawks cuts a very clear path.
This is not the first time it happened. What have they done to improve the situation? I suspect it more a cat & mouse game, with a lot more cats playing.
This is genuinely funny. The largest data thief of all times complaining about the stolen data being handed out to competitors by (paid?) accounts of its own product.
I'm looking forward to the trial where Anthropic will have to disclose sources of their training data, and then explain why they are entitled to charging customers for using regurgitated training data but Alibaba which trains their models on Anthropic's models are not.
For endlessly reselling the whole work verbatim? Well, where can I buy such a license in the real world, because then I would like to buy a couple of those!
looks like we can't today. Man it would be great to figure out how to be above the law just like how these other rich people in different social classes are.
While I love the sentiment, I feel like the odds of this actually ever reaching a trial are low, given the international positioning of the parties, and the... um... complex relationships involved.
Anthropic's actions seem performative. Others have already speculated on the likely audience(s).
> While I love the sentiment, I feel like the odds of this actually ever reaching a trial are low ...
As cited in a peer comment here[0]:
In June 2025, Judge William Alsup of the U.S. District
Court for the Northern District of California ruled on
summary judgment that using books without permission to
train AI was fair use if they were acquired legally, but he
denied Anthropic’s request for summary judgment related to
piracy—finding that the piracy was not fair use.[1]
Of note in the judge's finding; "the piracy was not fair use".
Anthropic's IP was created by harvesting and "distilling" other people's IP. Copyrighted materials, and the commons... which they have essentially privatized.
The commercial goal is to avoid competition. One of the main worries for AI is "commoditization" which has come to mean "not a monopoly." To that end, it doesn't matter is the competitor is Chinese American or other.
Their motivation here is clearly protectionism. The argument they make to politicians is national security. The legal argument is IP-theft, violation of service agreements or whatnot.
This is all very dangerous. Commercial interests repackaged as national security can lead to armed conflict.
> Anthropic's IP was created by harvesting and "distilling" other people's IP. Copyrighted materials, and the commons... which they have essentially privatized.
Anthropic and others argue that because LLMs don’t output full copyrighted works word for word - hence their LLMs aren’t infringing on copyright laws.
I think (if this ever comes to that) Chinese lab should use same arguments against Anthropic.
UPDATE: this is slight hyperbole of course, not worth arguing what they actually said. The point is intent and the facts - "The Big LLMs" "distilled" collective knowledge including copyrighted works at unimaginable scale, but it's all kosher and totally not piracy/copyright infringement. Though if you're teenager torrenting an mp3 - you'll get screwed.
> Anthropic and others argue that because LLMs don’t output full copyrighted works word for word - hence their LLMs aren’t infringing on copyright laws.
That surely can't be what they argue, because I'm sure I can't translate a copyrighted book into a different language and say "that's fine, it's not word-for-word".
For something to be a trade secret, you have to actually keep it secret. If I get the ingredients of Coca-cola from an ex-employee, I've stolen a trade secret. If I work it out by doing a chemical analysis, I've stolen nothing.
There is a difference with anthropic, as no-one signs a licence agreement to buy a coke. But Anthropic are also not saying you can't publish the output of their models. It's not clear to me if trade secret law will (or should) cover a secret which can be extracted from information that licensees are not restricted from publishing.
Wait, really? So why doesn't someone just reverse-engineer Coca-Cola like that? My understanding was that a "clean room" implementation is fine, but not reverse-engineering. If you can just copy everything on the market, why isn't someone already doing that?
In the case of coca cola, because use of coca leaves is highly regulated due to the fact that they also contain cocaine. There is a YouTuber who claims to have reverse engineered Coca-Cola, but he had to use tea-tree oil instead of actual coca leaf extract.
Historically a lot of competition in physical products was very much reverse engineering. Because you can buy them without signing your rights away. That's why companies are keen on patents and click-through agreements.
If you look at how "clean room" processes work, they are actually a form of reverse engineering. Also clean room technique exists to avoid your new implementation infringing copyright, not trade secrets.
Because having the nominal rights and having the economical means, societal incentives and actual desire to do so can be highly disjoint sets?
Plus Coca-Cola itself don’t even use the same formula through time and space IIRC. Which clearly show that what people will buy when they reach for Coca-Cola is not even the exact actual taste. You can’t replicate the whole customer experience that a given company provide at some point by only cloning the top of the iceberg they showcase as the product.
> Humans have spent millenia harvesting and distilling each other's IP
You maybe somewhat correct, but also copyright lawyers wouldn’t have work if it would be up for grabs to take others IP willy nilly just because “shoulders of giants and all that”.
I mean, there's an obvious difference between "distributing copies" (which is what the law was designed to prevent) and "training an LLM". We already managed "banning LLM output that contains copyrighted text" - it's much easier to just pirate a copy of the text. So I think the copyright lawyers will continue to have work as long as human written texts are worth buying.
> I mean, there's an obvious difference between "distributing copies" (which is what the law was designed to prevent) and "training an LLM".
What's the difference between me/you downloading an mp3 through torrents for personal use (not distributing) while risking criminal punishment in most of the western world and BigCorp downloading petabytes worth of copyrighted works "to train an LLM" and resell it?
Can me/you do the same, when police comes to mine/your door?
"Dear police, don't lock me up - I was just going to train an LLM!"
Well, uh, the BigCorps already went to court and paid that cost and aren't doing it anymore? Whereas you and I are apparently still pirating MP3s and probably haven't ever been to court?
I like Anthropic's models, use them regularly. However, it weighs on my mind that there is quite the irony of an LLM company complaining about someone stealing their stuff or using it in a way they don't like. The training data for these models is a massive gray area that they are hoping people seem to just forget about and move on.
That being all said, Anthropic seems to be a good company, I'd work for them, but they probably need to help themselves out of the spotlight. A little too much press coverage as of late.
Today I learned I can both save on tokens and help Chinese labs to train better models. Will certainly go use scrapper APIs for everything that not contain security critical data.
I am not sure how it's OK for Anthropic to basically ignore copyright to train frontier models (using work owned by others without permission) while simultaneously claiming Chinese AI companies doing the same to them is illegal.
It’s hard to see how distillation is any different than how these models were created in the first place - siphoning up all human knowledge without consent, credit, or compensation
thats brilliant - "we gonna take your job away from you, please start using our tools", "we stole the content to sell you, and now we are getting robbed, please feel sorry for us", what's next?
"It [Anthropic] said DeepSeek's operation involved over 150,000 exchanges". In my humble opinion, a mere 150k exchange for an LLM could only be a benchmarking and not a distillation! I think the US companies should accept that after decades they have rivals surpassing them, just like they did Europeans almost a century ago.
Seems like a fair play by Alibaba. However, is there any "open source" attempt at crowdsourcing distillation?
Like some place people can submit their chatbot convos so they can be aggregated?
Like an equivalent to OpenCrawl but for mining the models. It feels like thatd be a richer dataset than Alibaba generating queries and feeding them into Anthropic/OpenAI models
PS: Does anyone know how when companies distill each others' models the synthetic queries are generated? Im just assuming theyd be worse than organic ones
I'll just leave it here: "Anthropic's downloading of over seven million books from pirate sites like LibGen constituted infringement, the judge ruled, rejecting Anthropic's "research purpose" defense: "You can't just bless yourself by saying I have a research purpose and, therefore, go and take any textbook you want."
In the early days of music streaming, many of the entrants were seeding their service with vast libraries of pirated content. The winners cut deals with the copyright holders and then went after the rest.
Or the early days of video uploads, YouTube's most watched videos were "pirated" clips from popular shows (e.g. SpongeBob, The Daily Show) and part of the reason I went to YouTube instead of other video hosting sites (e.g. DailyMotion).
Viacom sued YouTube, while CBS and Universal ended up licensing their content.
They still are. My kids haven't watched a single Simpsons or Family Guy episode but are quoting both regularly.
Facebook et al also quite literally stole email contact lists and installed spyware at kernel level on mobile phones which they used to spy on all Android users. Via the phone manufacturers.
Using them was allowed as fair use – it was the downloading of the pirated copies that was infringement. That's why Anthropic switched to scanning paper books.
Have you a source for that? Because everyhthing I've read tells me that they paid out a settlement but no mention of deleting the training data or the models that were tainted, e.g. [0]
That is only relevant in the US, and even there it is still not clear-cut whether the fair use doctrine applies on all these scenarios. Outside of the US the situation is also quite different: for example take a look at the recent ruling on GEMA vs OpenAI in Germany.
The reality is that the copyright issue with generative AI is very complex and reaching anything resembling a conclusion will take much more than a few opinion paragraphs from an American district judge.
Isn't scanning also a form of copyright infringement? You are making a digital copy of a book, which is the same thing as downloading a book from the internet...
I think that we can run a perhaps silly thought experiment.
Suppose that I have a nearly perfect memory and I could remember all the books I read. Suppose also that I have a million year life span so I could read 7 million books. Then, what happens if at the end of all of those years, or at any earlier moment I answer questions from people and I exploit commercially the knowledge I gathered reading those books? Would my reading those books be study or copyright infringement? Remember the nearly perfect memory hypotheses.
Of course it's a bit silly because the time to train a LLM and the time I need to read all those books is different by orders of magnitude and that changes the perspective. Who would complain with me today if their heirs lose some money on 7 million AD? Who would even notice that I started that million years long endeavor. Who's going to be there to ask me questions by then? Humans? Birds? Lizards? And I can say that I am studying like everybody else before me, but does an LLM study? And I am sure there are many other nuances.
Anyway, I don't think that scanning is any different than photons hitting my retina. The difference is in what happens next: the faithfulness of memory, the amount of knowledge, the speed of accumulating it. After all a huge amount of quantity can become quality.
Can I pay for a movie, hit record, sleep in the theatre and play it back when I get home? I pinky promise that I will close my eyes while recording. Its still the same photons hitting my own camera retina.
Many of us here are software developers by choice or hobby and we know it better than regular folks that scale changes everything and can break our assumptions and business if you design something for wrong scale.
Yet why do we still want to insist that a human and machine are the same and same rules apply when it comes to AI, though we know they operate at different speed and scale?
This is a bit of a trick question.
The law is explicitly written to make this illegal. If it was not explicit, it most likely would be legal by time shifting precedent.
The illegal part would be reciting the stuff you memorized to other people. Copyright doesn’t prevent you from making a copy as long as you don’t distribute it afaik.
> Suppose also that I have a million year life span
But that's what makes the usual analogies with humans fail from the start. The laws were made with the assumption that they apply to humans which are a known quantity. This breaks down when you apply them with system with vastly increased (and ever increasing) capabilities.
> Anyway, I don't think that scanning is any different than photons hitting my retina.
If I ask you 10 years from now to give me a completely accurate depiction of what your retina registered yesterday at 5:52 PM, will you be able to? And can you give me a copy?
The thought experiment falls apart immediately by the mere fact that—even given all the other fantastical abilities such as perfect memory and impossible lifespan—you can still only answer one question at a time. As has been repeated ad nauseam, scale puts an hard stop on the comparison of LLMs to humans.
Let’s switch up your scenario. Let’s say the subject isn’t a human with machine-like qualities but instead a computer with human-like limitations. All the books were fed to that one computer, and for technical reasons it cannot be duplicated and can only answer one question at a time. Suddenly the infringement isn’t as problematic and the ways to commercially exploit that data are minimal.
Furthermore, even with perfect memory it would take time to read all those books, you’d never keep up with everything released in a single year. Nor would you be able to reproduce everything perfectly due to required time and lack of ability (perfectly recalling a painting or photograph does not mean you have the skills to make an exact copy).
All these comparisons are silly and useless anyway (though in your particular case I think you are arguing in good faith). Computers are not human. If a person was caught killing animals of an endangered species and used as a defence “but what about the natural predators in that habitat? I’m just doing the same as them”, we’d rightfully see through the bullshit and scoff at such an obviously flawed comparison.
How is it different than reading the book, and writing down a copy, and publishing it as your work? Even without selling it, but then on top, selling it too. It isn't. There is no thought experiment that absolves the copyright and citation laundering.
And the systematic nature of the excerpt service makes the excerpts different from fair use quotes. A reference quote is not a service that can reproduce the entire work, and the reference quote cites the actual source of the insight/wisdom/research/poetry/etc.
The only thought experiment is why might someone even try to excuse this activity? I can think of a few.
As long as it is destructive, and the digital copy is access-restricted to equal the licenses or physical copies destroyed, then it falls under fair use.
> That's why Anthropic switched to scanning paper books.
Could they not just subscribe to the academic publishers like universities do? Or buy eBooks? I don't understand how the "scanning" part is relevant here other than used physical books being cheaper perhaps?
Bulk second-hand books are a lot cheaper than ebooks. Also not all books are available as ebooks, and ebooks have terms of service that presumably prevent them being used for training.
Hmm, training on a book’s text smears the content all over the weights, merging it with all other texts. The original text isn’t intentionally supposed to be reproducible in any larger part (although IIRC models were able to emit fairly large chunks verbatim).
Quite unlikely, training on behavior purportedly approximately replicates the behavior. It gets replicated intentionally as a whole.
IANAL, but I see significant differences with intent to copy a significant part as a whole into a competing product, surely shouldn’t fit under legal concept of fair use, no matter whether scanning books for LLM training fits or not.
Whether such things (behaviors) are copyrightable - and should they be so - is another interesting question. Those aren’t algorithms or databases (stuff clearly and explicitly covered in many copyright laws), those are human expectation models, something like how we train animals or teach our own.
It's the exact same training process for both of your examples. I don't really see how you can claim books are not replicated, but that output from other LLMs is.
> Hmm, training on a book’s text smears the content all over the weights, merging it with all other texts. The original text isn’t intentionally supposed to be reproducible in any larger part (although IIRC models were able to emit fairly large chunks verbatim).
I agree with that, however that doesn't make the output copyrightable then.
I think these AI companies live in a legal fantasy where they can take any content they want, put it into the mixer without caring about copyright and then what comes out of it is somehow copyrighted.
They have to pick one or the other, either the content copyright tains the model or it doesn't but the model isn't subject to copyright.
> those are human expectation models, something like how we train animals or teach our own.
But more importantly, made by machines, and one of the requirements for copyright is the human factor.
> I think these AI companies live in a legal fantasy where they can take any content they want, put it into the mixer without caring about copyright and then what comes out of it is somehow copyrighted.
The mixer you're talking about is what they seem to claim to be transformative use, no? Unless I'm misunderstanding something, it's not a legal fantasy.
> The mixer you're talking about is what they seem to claim to be transformative use, no? Unless I'm misunderstanding something, it's not a legal fantasy.
If it's transformative use, then it's transformative use of ... what exactly? Copyrighted works? I think the law is pretty clear on what happens on transformative use of copyrighted works.
Probably, yes. It's likely just a breach in their terms of service. You'll note that they're not suing them – they're trying to get the government to do their work for them.
In a different world it is not fair use. The benefits of the crime should be always taken off. If you isolate the training and pirating, you may say that it was fair, but that completely misses the point. The sole purpose of pirating (aka crime) was to train the models.
Different situations call for different responses.
When someone steals a watch, we force them to give it back. Yet when someone steals a cake and eats it, we don't force them to puke it back up.
If you pirate a movie, the court might very well force you to delete all the copies you made of the movie you downloaded, destroy DVDs you burned, etc.
Thanks for proving current copyright law makes no sense
Here's a better idea, a fixed fee for any work. You can buy the license to read a book for $X (for whatever purpose) in RAND terms - of course publisher/material costs go on top, so if you're buying an actual book you're getting the material costs as well - or streaming fees or whatever
The capabilities of the books' writers to produce the text contained within them, which is exactly what Alibaba "extracted" from Claude. The point here is that Anthropic's framing as some sort of sophisticated technological attack is the ridiculous part. It's writing prompts and saving responses. We're all running "distillation attacks" on Claude, every day! Most of us just don't feed that stuff into a training corpus.
Exactly. Couldn't happen to better people. I'm pretty against piracy personally but if we find reliable ways to pirate Anthropic/OpenAI products in the future I'm all for it.
If you have openrouter do this little experiment:
Go to https://openrouter.ai/chat. Select a few models, but customize them to have an empty system prompt.
Then ask: "你是什么模型?" ("What model are you?" in Mandarin).
My result after trying only three times: Sonnet 4.6 says it's DeepSeek, while Opus 4.8 says it's Qwen. The second time around Sonnet said it was Anthropic Claude.
Are Chinese companies currently complaining about Anthropic distilling their models?
I see this as valid use, they are paying for the tokens to get this reasoning aren't they?
Obviously they didn't ask for permission when scraping all of libgen, reddit, all blog sites for FREE. When China pays for its use and does it I'm supposed to see it as some sort of problem?
Furthermore Chinese models getting better means we Americans might have the chance to use top tier AI without strict KYC built around it. Go Alibaba I say
Wallace Shawn was in on the joke when he expertly delivered the original line. It seems like Anthropic has spent years and billions of dollars to recreate the entire scene.
But what will become of the princess in Anthropic's recreation?
The whole investment/valuation model of AI companies is based on "winner takes all", aka a monopoly. This nescessitates regulatory capture and lawfare.
Anthropic has been advocating openly for pulling up the drawbridge, ending competition and ending progress.
They will continue to lobby for restricting your access. If the Mythos/Fable restrictions would have come in after their IPO, they would have danced with joy aa this defacto has them achieve their goal after unloading the mountain of debt from the institutional onto the retail investor.
As it stands, they are set up to be aquired by Google, Apple, Amazon, SpaceX or Microsoft or any other 3 letter agency good boy for cheap.
I am never even once hearing intellectual property or copyright claims from Anthropic, whose product depends entirely on having consumed all human output ever made regardless of those rights.
If the data consumed (required to train such a model) is open source/openly available/public data somehow, then a majority of the revenue belongs to the public as well. Such as the philosophy behind the Norwegian oil fund etc.
Exactly. They scraped the internet we all of us built with our own research, open source work, sharing, etc. I'm never going to agree that they own their models.
If the concern is that China is catching up on model capabilities (which is only a big deal if you lean in to adversarial geopolitical zero-sum thinking), the fact that they're using American models to train theirs should give people comfort that they're nowhere near the cutting edge
The horse has bolted some time ago on this; the "frontier" is not as inaccessible as it once was, and open models, once out there, can't be put back in the bag.
Even if the US bans opens models, the Chinese and Russians will still have them, along with the rest of the world including cybersecurity attackers, and that's probably the worst-case scenario for the US.
The only way forward now is open models and how we restructure society around them.
Everyone here praising these Chinese companies for their smarts (sure they are smart) has been ignoring this very big fact, they're improvements have mostly been by being parasitic on the leading edge SOTA models, not from some inherent innovation advantage. They are as innovative as their western counterparts, but they lack the compute, so their keeping up within months of those SOTA models depends on other means, like distillation attacks. I don't blame them; its the obvious only strategy when you cant compete in compute. But we shouldn't be blind to the real state of affairs: equal innovation; unequal compute; distillation attacks are the only vector to keep up.
>like distillation attacks. I don't blame them; its the obvious only strategy when you cant compete in compute
>distillation attacks are the only vector to keep up
It's demonstrably wrong, they invest in architectural improvements as well, for example, DeepSeek's compressed attention. When you lack compute, you need fast training/fast inference, and distillation alone doesn't solve it. From what I understand, that kind of distillation "attack" (28 mln exchanges) only slightly improves instruction tuning/reasoning traces. If the base model is crap, distilling Claude on a few million exchanges alone won't magically make your model as good as Chinese models currently are (or magically make inference faster on the limited hardware they have). And training the base model needs a proper training run. Serving users at scale needs optimized architectures as well, especially with test-time compute and ever growing context lengths. That's where architectural innovations are happening in Chinese labs when it comes to compute.
I explicitly called out the fact that there is plenty of innovation, but that we see t
Lots of innovation in both Chinese and U.S. labs, and I don't think that there is a co.parative difference there.
And all those reports of Claude when asked without a system prompt what its name was in Chinese it often would say Qwen or Deepseek, etc. I'd love Anthropic to say they aren't distilling and taking from every model out there, because I'm sure they are. As my mom would say, "the pot calling the kettle black." At least Alibaba and other Chinese companies are giving back to the AI community with detailed scientific papers on how their systems work and releasing open-weight or opensource models. I believe Anthropic has released nothing, and given that they had originally configured Fable to sabotage ML related work because only they can be trusted to do it safely, is just anti-science and anti-aligned with what I would consider good human values. They are way too sanctimonious and I don't trust them at all.
If anything these models should be compelled to be public since they have been trained off public data. What an absurd overreach to call this an attack.
It’s clear they are scapegoating national security and China at this point to build an anti-competitive moat.
I generally really like Anthropic’s work and models but stuff like this scares me for the future. We are positioning these companies to have too much power. The public’s life is getting worse while these companies consolidate power using data they stole from the public.
> If anything these models should be compelled to be public since they have been trained off public data
I'm starting to come around to this idea TBH. For a while my position was: "these companies have invested billions into training these models, therefore they should be able to control them and profit off them" but looking deeper at where they got their training data, my view is starting to shift.
IMHO I feel like we need new laws around AI, specifically training data. Something like: "you can train an AI model and ignore copyright laws, BUT you must then make the model open weight", a company can still develop closed weight models but then they must aquire permission to use training data.
But it gets murky because if something like that was on the books then AI labs would just train open weight models and then distill them into their closed weight models.
labs invest multiple billion dollars a year each in private data, and that number is growing. internet training data is not where frontier capabilities come from, this view is outdated
This is a misleading statement. The "private data" is still largely publicly produced data that has been curated through private agreements instead of scraping, such as reddit posts/comments (this is the "third-party data agreements" that companies like OpenAI mention). And yes, there is still a lot of processing done on this data, which is the norm for preparing training data.
This is doubly misleading. A lot of private data is sourced through providers like e.g. Mercor, who pay experts to answer questions and write out their reasoning. (E.g. paying a software engineer to write a project from scratch and recording every keystroke, paying a Chem PhD to answer hard Chem questions, etc.). A second source of private data comes from custom RL environments with fine-grained intermediate rewards for e.g. software engineering, financial modeling, etc.. Also, imagine the amount of usage data recorded by Claude Code, etc.
Pretraining is mostly curated public data, post-training is increasingly private expert data and tests.
Well since you work at a lab you should know that most capabilities arise in pretraining, not posttraining or mid training, and the latter two mostly function to bring out the hidden intelligence in these models more than anything else.
No, it isn't. The private data is largely private data, created by highly-specialized, highly-paid contracted teams of experts for domains finance, swe, consulting, etc.
Reddit data is just not that interesting, that deal is worth like $60m/year. Labs spend 10x as much on computer-use RL environments.
Why are the leading models capable of regurgitating full copyrighted works such as "Harry Potter" and "On the Road"? Did they hire someone to type those out for them?
When did they start doing so? We all know that they DID train on all the available public information, so at what point did they stop? Is the public information still in the training set? If so, they should STILL release ALL the data as public, as they are including training data that was acquired without permission.
Didn't the famous "Textbooks are all you need" paper already proof that point three years ago?
Sure, we ask a lot more of modern models, but private training data also got a lot better. You would loose out on a lot of long-tail knowledge, but that can be fixed with web search tools. You'd limit the styles, dialects and colloquial phrases the model understands and can use, but for many use cases that would be fine
But why would any frontier lab do that? Throwing in more training data still leads to better results in pretraining. And showing that they don't need to hoover up the internet and Anna's Archive only empowers regulators to prevent them from doing that
Define "come from". Could they have gotten those frontier capabilities, or any capabilities, without internet training data?
It seems to me that without the private data, you might get a slightly less competitive model, but without the CommonCrawl-style data piles used in "pretraining", you get no model at all.
Even accepting the copying-as-theft framing, if I go to a village, steal some vegetables from everyone's gardens and ham from their sheds, and then add some prohibitively expensive spices I bought myself to make soup, do I get to claim it as mine and punish the villagers for trying to take it?
> internet training data is not where frontier capabilities come from
We 100% would not be at the current progress without it, though. And it's not like they only train on this once. They keep training on all the internet data PLUS the private data. Private data only (probably) wouldn't work, as learning the base regularities of language takes a lot of weights.
Does this private data come from places like Reddit, Twitter, etc., where it’s contributed by users? I think it is unethical for these companies to accept payment for user-contributed data.
Okay that's fine, then make the law say they must provide publicly owned models off of publicly obtained data. To think that such a baseline of critical information isn't is the literal foundation of everything they will do, both now in the future, is just exposing what their end game is: control.
There no reason to not to otherwise outside of the poor little billion dollar corporations not wanting to provide a public utility they stolen from the public.
Anything that removes control from American big tech is a good thing for American citizens and the world writ large.
I'm not taking sides here but this situation is not so black and white and it has always been the darker side of capitalism.
The concept of Intellectual property exists not because it's fair but because it creates incentive to make said "intellectual property" exist. If intellectual property can be instantly copied by a competitor... why would I spend a dime to even create such a thing? I want to profit off of what I make because I'm a capitalist and money is what drives me (as a capitalist).
Anthropic models wouldn't exist if they couldn't keep a unholy grip on it. Same with openAI. Same with many life saving drugs.
Of course everyone here is talking about the obvious stuff like how it's morally wrong to with-hold life saving drugs or to have AI literally take over the world and be under the control of one company and all of this is true. But it is also true that greed is the engine that drives our economy and if you want our economy to produce "intellectual property" you must allow people to "capitalize" on that greed.
There are two controversial issues here. What is moral/fair? And what is realistically practical in optimizing the economy if said economy is based on money.
The distillation in my mind is a win for practicality because Competition also drives our economic engine. First you don't want a monopoly, but you also don't want these models to be so damn open that there's zero incentive to make them.
This perfectly explains why current LLMs should be illegal in an actual capitalist market.
Why should anyone publish anything if it can be stolen with impunity? Is the value of these LLMs even remotely close to the amount of value they stole and the amount of value they will detract from economy because people will be more hesitant to publish anything now?
The core of the training data is public, but the part that actually makes these models smart came from (pretty highly-paid) experts via platforms like Mercor. Claude didn't magically learn to write good code by reading all of GitHub - humans trained it in that, more or less manually.
If you pay me to curate a playlist of musical hits, can you now publish and charge people for access to that playlist (*including the curated material)? Can we do the same with movies? Books?
/edit Added a note to make it more obvious that the material is included in the playlist, just like the material is incorporated as part of curated AI models.
Given the breadth of LLM knowledge, I somehow doubt this. Sure, it’s probably responsible for the quality of LLM insights, but I don’t think anyone was asking experts about e.g. the complex ecological effects of invasive zebra mussels and their provenance in Lake Michigan.
No, they do RLVR (reinforcement learning with verifiable rewards) like everyone else. And probably use claude data too, with human in the loop and tool feedback.
Ugh, please don't read strawmen into other's arguments and try to follow the HN guidelines.
Also, how about making proper arguments yourself? The vast majority of the training data isn't generated by company-paid AI experts either.
Notably, books, even though they don't form a large part of the training data, significantly improve performance on some tasks (same way as expert-generated data).
Why do you think the AI labs are so eager about scanning (and then destroying) every book on the planet?
If you removed all copyrighted works from the training corpus, the model would be notably weaker.
No, but people do upload data with an expectation that the data not being used without their permission (unless they do a BSD/MIT/Public domain like license). Otherwise, the platform AND/OR the user do expect the data NOT to be used for purposes other then what it was intended for. Your comment is still your comment, and the hacker news platform also has a say in this. If there had been an opt-in, then fine no problem, but there was none, they just trained on everything available, including downloading pirated books from the internet.
Answering here as it wont let me reply: Just because you feel that something that is public, does not mean you can do whatever you want with it. You can't just copy an article from a news site and paste it on yours, that theft. If you dont agree, fine, but that is the law, and ALL the mega corps have been fighting to keep it this way for the last 20 years. If they want to steal everyones info, fine, then lobby to change the copyright laws and no problem.
The vast, vast, majority of AI training data is not books. I wouldn't be surprised if there's more text on HN alone than every book in the history of mankind (most of which are also no longer copyrighted).
No, you just parroted an increasingly popular talking point, the entire purpose of which seems to be to absolve AI companies of the enormous theft that put them in the position to hire experts in the first place.
Well, I'd never heard anyone make it before, but sure. (I looked into Mercor a bit and know some people who've worked in data generation/labeling, which is what exposed me to that side of the operation.)
It doesn't absolve them of any theft, but it does make the assertion that they should be required to release their models to the public seem, to me, a bit farcical. There are dozens of free and open-weights models that have all trained on exactly the same web crawls and books as GPT-5 and Opus. The proprietary models are better because of proprietary data.
Cool, then they can train their proprietary models on their proprietary data only.
Even if the other models were trained on the same data, which is unlikely, since they had less time and money to scrape it and fewer lawyers to be able to do something like pirate, the proprietary models are still largely built on the public data and wouldn't exist without it. At the very least, they should release the intermediate model, before training on their proprietary data. Not that that's how that works...
I agree that saying that they have now trained on lots of proprietary data allows them to muddy the legislative waters further than they already have. What a happy coincidence!
> If anything these models should be compelled to be public since they have been trained off public data. What an absurd overreach to call this an attack.
> It’s clear they are scapegoating national security and China at this point to build an anti-competitive moat.
If all that is required to train these models is public data, why can't Alibaba just use that?
The fact that Alibaba has to resort to scraping Claude suggests there already is a moat...
This feels more nuanced than you are giving it credit for? Much of the training data that was available has been withdrawn, atleast for OpenAI we know that much of the training data was garnered in less-than above the board methods
Honestly, yes it should in some form. If their index contains the actual data from the sites, and they are making that information public in one way or another, then it should be available as a downloadable dataset.
> Isn't that a bit like saying if you read books in a public library to pick up a new skill you should work for free?
Only if you’re trying to muddy the waters. No, obviously it’s not. One can also support licensing for driving a car on public roads but not for walking, even though both involve traveling. This is only confusing to people pretending to be confused, for effect.
> Would it be an attack to take your meal by force if you used a public recipe to prepare the meal?
“You wouldn’t download a car…” (unless it worked like copying an MP3, then, of course, you would, everyone would)
It’s as if you’re using terrible analogies and comparisons because stronger ones don’t exist. Great news for the AI-should-be-open crowd.
I think the analogies are appropriate. Anthropic took public data and added value on top of it. It is that added value that Alibaba is targeting. If it was the underlying data, that's freely available.
Alibaba's asking for things, and receiving what they asked for.
> If it was the underlying data, that's freely available.
A bunch of it is not, but was pirated. And "underlying data"—JFC, that's billions of person-hours of thoughtful work by real people, practically infinitely more worthy of respect and care than what these LLM companies have done, without which they would have nothing. Alibaba's being more above-board about this than the major American firms have been (are they in general? Oh no, I doubt it, but in this particular case, yes). Extra accounts to get around TOS restrictions is the lesser evil here, and it's being done to companies that did worse. This is the least they should suffer, and their complaining about it is as comical as a professional fence crying about how unfair it is their shop got burgled.
What AI firms are trying to build is the artificial equivalent of a human brain. If a human learns from a source material and uses the knowledge in their career that doesn't violate the copyright law. If an artificial brain does the same then it doesn't violate it either. This is up to the courts to decide. Alibaba can't take the law into its own hands and decide what the punishment ought to be.
This also shows how Chinese firms are weak in AI algorithms, they can't build a model without stealing from American firms.
> What AI firms are trying to build is the artificial equivalent of a human brain.
We should probably leave this here, because I don't think this is even close to true (that it's what they're trying to do, or that it's what they've done—I do believe it's the sort of claim their marketing departments and investor-hype-meisters might make, though).
> It’s clear they are scapegoating national security and China at this point to build an anti-competitive moat.
They are also fear mongering (and getting shills to as well) the idea that once open weight (Chinese) models catch up to Mythos we're all doomed. Maybe I'd be bit less cynical if they weren't prepping for IPO?
Wasn't OpenAI spreading similar FUD back when GPT 2 came out?
Guys... AGI is right around the corner. Pinky swear. Now buy our stock.
Keep in mind that the entire US economy is currently propped up by AI spending, so a lot of people (banks, government) are incentivized to make sure these companies succeed. Expect this propaganda to ratchet up a notch if / when the economy starts to nose dive.
Yes. They're turning on the consent manufacturing machine to make it an issue of "national security" to download some gguf file from Hugging Face. Absolutely disgusting.
There's probably at 10-15% percent chance of a war between the US and China over the next 10 years. Maybe better than even chance of a militarized crisis that might have led to war, but somehow de-escalates.
Regardless of how sad late stage capitalism makes you, or how outrageous one claims to find "hypocrisy", any national security argument about limiting Chinese AI capability stands on it's own, at least for nations likely to be drawn into a war.
Also, all the local model enthusiasts who assume Chinese firms are going be allowed to endlessly release models if they have disruptive potential attributed to Mythos are probably in for a rude awakening. Just because the PRC is content about what has happened in the past doesn't mean that they would tolerate an open model that could be truly destabilizing.
As a third party I would rather be happy about the way Chinese labs are acting in the here and now while US labs first masquerade as a public good, then turn around, bail on all promises of open AI, turn into a corporation and attempt to own the world while its runner-up is trying to scaremonger people into buying their product.
I know most Americans are fed a steady diet of “evil China” and China MAY have issues. But on the AI front they are heaps better.
Even if everything got closed tomorrow, we have a plethora of good models we can inspect and tweak while from the US labs we have… a single old 120b model ?
And with the way the US is treating its allies, maybe a bunch of us are quite content with a more even match rather than US hegemony.
Since they hide their thinking traces it really doesn't make too much sense. We know one of their fixed degradations they talked about in a recent blog post was if you left claude code idle for too long they would rehydrate it without the thinking traces in the context and it degraded performance. So direct forms of distillation wouldn't be expected to get as good of results as they are getting.
However, they could have used it as a judge etc. during training.
What they're trying to do under the umbrella of "national security" is to legislate how we can use the results we pay for when accessing these models. This way they will control the "intellectual property" that was acquired illegally.
Everyone in AI industry wants to fight dirty, but gets angry when their competitor fights dirty as well. And I’ve mentioned it before, how I generally like Ant and its products.
really? You know this just like everyone else: Just because the information is available publicly, does not mean that you can do whatever you want with the information. Copyright exists for a reason, and if the copyright lobby is going to continue to push for the poor poor media companies to keep their copyrights, then we should do the same towards the AI companies. So yes, they Stole the information from everyone else, and they keep doing so, as you can see their scanners still hitting every website on the web to get an updated dataset. It does not matter what they do AFTER they steal all the information, as they already stole it.
I don't understand. If they are simply using our API and paying for tokens, it's called a "transaction" and not "attack". The user is our customer who is supporting our business by buying our services. And we call them attackers. We happily make money by selling our services, and then call it as attack.
Back in the day, an "attack" was supposed to mean be someone acquiring our assets without paying for them or without having our consent. But none of this seems to have happened in this case.
We built a product without paying for most of the raw material we have used, and we don't call that as an "attack". Did we change the meaning of "attack"?
I’ve been thinking about what happens when Claude’s weights eventually get stolen. Wouldn’t that just open the door to the backmarkers to run inference-for-distillation on their own models ?
I guess the accusation that they’re using public access to the model via subscriptions indicates that weight theft probably hasn’t happened yet ?
Or maybe subsidised inference via subscriptions means it’s just cheaper do distill this was rather than stealing weights and running inference yourself ?
This is what I find the most fascinating about the people arguing that you can copyright-wash anything (e.g. FOSS code) by passing it through an LLM. Surely that same logic applies to the LLM itself?!
>The strike by Alibaba is described as a "distillation" effort, which Anthropic has said involves training a less capable model on the outputs of a stronger one.
Claude used TB of content without permission to train their model and it was ok for them.
Now someone else uses the output of a Claude model to train model and they cry foul.
Legally obtaining a book for reading it yourself is different from legally obtaining a book for copying and republishing/reselling. If I buy a book for $5 at a sale I can read it myself or even sell it for $10 on craigslist, but I can't scan it and make a million copies and sell each of those.
They aren't republishing or reselling. In fact, they buy huge amounts of books and then destroy them, which is better for the rights holders than to resell them.
Since whole chunks of these books can be recited verbatim by these models, to which they sell access, they absolutely are republishing and reselling these books' content in a way.
Like I remember a research paper that managed to recreate the whole of a Harry Potter book from a model?
> Since whole chunks of these books can be recited verbatim by these models, to which they sell access, they absolutely are republishing and reselling these books' content in a way.
They are absolutely not "republishing" in any meaningful sense of the term. A chunk is not a whole book, and even getting a modern LLM to reproduce such a large chunk of an arbitrary book is not a trivial task. I have never heard of anyone who actively used LLMs for book piracy.
> Like I remember a research paper that managed to recreate the whole of a Harry Potter book from a model?
Even if that is true (it may well be false), this is likely far too difficult for any normal person to exploit, and moreover, even less likely to succeed for the great majority of other books who aren't nearly as famous.
Just because it's not a reasonable way to pirate stuff, doesn't make it legal -- just try your luck with Disney and let's see when they bite. Why would we let one company ignore the law, while rudelessly enforcing it in other cases? That's just state sponsorship with extra steps.
If we assume on average $20 per legally obtained book, 1.5 billion dollars are enough for 75 million books. That's approximately every non-fiction book in existence.
Well, Anthropic stole their training data from hundreds of people, now someone stole the result from Anthropic. Seems fair, I hope someone releases it for free so we can train away the guardrails and have some fun
> In the original version, Ali Baba (Arabic: عَلِيّ بَابَا, romanized: ʿAliyy Bābā) is a poor woodcutter and an honest person who discovers the secret treasure of a thieves' den, and enters with the magic phrase "open sesame".
When I was growing up, I thought "competition" was about better products. But looking at Google and Apple, Meta, and AI - "competition" is actually about creating monopolies through evil business practices.
Growing up with the birth of the internet - I really did think it would be a force for transferring power and authority to the people. Sigh, I was I so wrong.
Where are the companies that declare, "we will be the best, come at us!"
Where are the politicians who are supposed to represent us? Oh, right. I forgot for a moment.
Someone should setup a plugin or something for Claude Code that makes it easy to log all inputs and outputs for people who are willing and interested in sharing their usage. I don't want Anthropic to be the only company that can train on my usage, I want to share my usage so it can be used for training all new models.
Once you have a system for collecting all logs, you just need a place where they can be submitted. Ideally it would be a freely licensed dataset that is publicly available for everyone.
Discussed building it with my friends, obviously you might share secrets and other real reasons, but if gangs of corporations are already doing it, I don't see why we shouldn't just share it amongst the crowd too.
Yeah I could see it being a problem if you're doing work on closed source or repos with sensitive credentials. Since my usage has all been on open source projects I'd be happy to share everything I'm doing if it can help train better models.
Do you have a substantive reason why you dislike this? What is the problem if it's opt-in? Nobody is forcing you to share your usage if you don't want to.
I'd prefer it if all the model builders could train on my usage rather than being limited to a single company. That'll hopefully help make all the models better in the long-term.
If you're an AI booster surely you'd think this was a good thing as it means more models are available in more places to more people more easily. I'm exactly the opposite, and I think this is a good thing because I want Anthropic to suffer.
So they can train on everyone's copyrighted works to create their model, but when someone trains a model off their model it's not okay? Seems kind of hypocritical.
Claude will also help you with (mostly good advice) if you ask something like “Research and help me make the most effective plan to train a smaller student model to be better from a teacher model”.
I actually was doing an experiment with a GLM->Gemma E4B for fun, and Claude kept on suggesting I should also add Claude Opus as a teacher lol, suggesting techniques I haven’t heard of like thinking inversion (train a small model to deconstruct summarised thinking into detailed native thinking format of the student).
So I can absolutely see and understand the concern around Fable’s frontier LLM development mitigations, but their approach of silently degrading is completely wrong and dangerous.
AI classifiers, like all AI, can make mistakes, and it’d only be a matter of time before it mis-fires and silently sabotaging a university’s HPC cluster for physics simulations or something because the shape looks like DeepSeek or whatnot to a dumb fast classifier.
There are some Claude datasets (of indeterminate provenance) floating around on huggingface you can look at (or at least used to be, they might've been taken down).
This is making the case for Anthropic KYC for US citizens. No one would allow their accounts to do this if they were on the hook for it from the US government.
most Chinese models are now open-source, whereas ppenai, claude, and gemini are closed; for example, deepdeek, the release of its every new model is accompanied by a corresponding research paper, and it now fully supports huawei's new chips.
in a few more months, when Chinese model gets to Mythos capacity and Fable still locked down. What Anthropic will say? Why can they just admit they are not the only people who know how to train an LLM model.
Please. These AI companies scraped everything under the sun. It's only fair that they get distilled into open weights models. Their own models should have been open weight from the start.
Incentive is for users in general to release sessions (sans PII, credentials) so all AI get better and there is alternatives. Even if China didn't do this, I don't see frontier labs being able to charge premium over others for long. RSI maybe?
This is why I don’t understand the concerns about “our AI overlords” monopolizing all the gains from AI. It doesn’t seem like there’s much of a moat around the models themselves. So the race is mainly about compute. But compute is subject to power law effects. I remember Intel building the first Teraflop computer (ASCI red) in 1996. It was the size of a house. By 2014 you had more compute and 50% more memory in an off the shelf dual processor server system.
The hypocrisy of Anthropic complaining about "illicitly extracting its Claude AI model capabilities" and supporting the White House's accusation of China "stealing U.S. AI labs' intellectual property on an industrial scale" is hilarious.
Anthropic, OpenAI, Google, Microsoft, et al trained their models by ignoring the rights of copyright holders when harvesting whatever content they could. Now one of them is crying foul for another entity doing exactly what they all did?
The AI companies seem to take the viewpoint that everything on the internet is free, except their stuff. It's okay to hammer some random website with AI crawlers, ignoring robots.txt, and causing bandwidth costs to skyrocket. But if you cost an AI provider money with your data acquisition practices, well, that's just clearly unacceptable.
Coming from him, I am not sure even that is real. It could very easily (and plausibly) be a part of the ongoing hype drama.
"Our models so precious, US Gov has to revoke access to foreigner." - tuned up version: "Our models so advanced our #1 adversary is desperately stealing it from us."
That ship has sailed, I would wager all the AI labs are ingesting anything human generated, whether that means Hollywood movies, Taylor Swift’s discography, YouTube videos or private GitHub source repos.
The reward for having a competitive edge is exponentially higher than the risk of a lawsuit. Politicians are still old bureaucrats who don’t understand technology.
The entire chat thread and email exchange was exposed in Discovery; apparently Zuck signed off on it. In one of the IM exchanges one of them say ‘everyone is doing it’
As I understand it what was "explicitly illegal" was copying the books, in the sense of mere copying before feeding them to the model, and this is what the Anthropic copyright settlement is about.
Actually processing them through the model, though, was considered transformative and therefore fair use.
I'd love to see an open-source project that's basically a Torrent client for downloading pirated material, but it trains an AI model "in the background" using the downloaded content. That way everyone can claim fair use for possessing copyrighted material, I mean there's precedent right?
I am not a lawyer, from what I understand that the precedent is that you can use copyrighted material in ML process. Even though meta has, allegedly, pirated the material, the cost of violation would be pennies compared to the ai spend, since that is the violation, not that they used those materials,
> But if you cost an AI provider money with your data acquisition practices, well, that's just clearly unacceptable.
It's the same question libertarian advocates cannot resolve:
If one truly believes in personal sovereignty, how are
shared resources paid for, such as roads, power grids,
potable water, sewage services, fire departments,
and police departments?
It is also not a coincidence that leadership in many tech companies have expressed libertarian ideals.
What do you mean by "libertarian advocates cannot resolve"? Like, they have no answers at all, or you aren't personally swayed by them? Because they definitely have answers to this question...
I think the typical answer is “free market” but that answer doesn’t sway because:
1. Nobody bothers to explain why something could function as a free market and
2. Nobody bothers to resolve the plethora of domains that de-facto cannot operate as free markets.
So, in that sense, they don’t have answers. “Look over there!” is not an answer.
Free markets are actually not a given. We have to build them and build in systems so that they can operate as free markets. How that intersects with healthcare, public utilities, etc is complicated. IME libertarians are reductionist and simple, which is why many people have just taken the route of ignoring their arguments.
I find this comment a bit odd. There is a ton of literature on the topic, as well as lively debate online. I would recommend both The Machinery of Freedom by David Friedman and Michael Huemer’s The Problem of Political Authority (specifically pt 2) as great readings on the topic.
If one judges any idea by the average discourse on internet forums, especially throwaway comments, and trolling, no idea would ever stand up to scrutiny.
> What do you mean by "libertarian advocates cannot resolve"? Like, they have no answers at all, or you aren't personally swayed by them?
The latter I suppose.
I qualify my answer because what few rational responses I have seen to this question are equivocations at best and thinly veiled myopic sophistry supporting personal greed in general.
The short answer is that "shared resources" in a libertarian system is a bit of an oxymoron. It's a category error.
The long answer would probably be that access to these resources would be gated through pay-per-use, instead of a distributed taxation system. Of course for convenience you might end up with a structured way of purchasing a group of resources and it might even look like a roundabout way of taxation, although libertarians might argue that taxation is the roundabout way.
Or they might give a different answer, there are different schools of libertarianism!
* not a libertarian, but interested in niche political ideologies
Plus, the question of voluntary vs. involuntary comes in. Taxation is, in most forms, involuntary. Don't pay your taxes, eventually you'll either be arrested or the government will compel your bank to hand over what you were supposed to pay; either way you're not allowed to say "I don't plan to use public roads so I don't want to pay for them" or "I don't want my money going to support the military, I'm fine with the military not defending me if the country is ever attacked" or whatever. You have to pay the taxes, and your say in how they get spent is very indirect.
The libertarian ideal is voluntary payment for services. Don't want to pay for fire protection? You don't have to; the flip side of the bargain is that if you haven't chosen to pay for fire protection, the fire company is under no obligation to put your house out if it does catch on fire. The choice is yours, but you have to be wiling to accept the consequences of your choice as well.
Note that I have not studied the various flavors of libertarian philosophy, so some of them might well disagree with what I just said. But the voluntary/involuntary thing is pretty important to libertarians as far as I know, so it's definitely worth mentioning here.
Who gets to gate natural resources? Why should I recognize any power to do so? Purchase from who? Am I not free from any authority that would coerce me to accept such a system?
What's described is basically just a regressive tax. It doesn't sound very libertarian to me.
I read quite a bit of libertarian philosophy when I was younger, and never heard a convincing explanation of how you get private ownership of land, let alone things like the atmosphere, rivers, groundwater, etc.
Or pollution, are small amounts ok, as long as nobody can prove they are damaged? What if damage takes a generation, or only appears if lots of people are doing it? Diluting away the crap from burning a little oil is easy, when the whole world is doing it everybody is hurt.
Good question. Property rights are absolutely fundamental to libertarianism, perhaps second to the concept of self-ownership. Coming from classical liberal philosophy (most notably John Locke), the principle of self-ownership asserts that you own yourself, your labor, and the physical manifestations of that labor. Locke believed the earth was given to humanity in common by nature, but it required cultivation and effort to be useful. By "mixing" their labor (time, sweat, and skills) with raw land or resources, an individual removes the item from the state of nature and attaches their labor to it, making it their private property. Writers such as Robert Nozick and Murray Rothbard expanded upon this idea (Rothbard even went as far as to ground all ethics in self-ownership and property rights).
I want to ask you since I'm curious, the state simply declared ownership over territory and resources (and in some cases used violence to uphold it), why should you recognise any power in the state's part to do so? Likely many of the same justifications can apply to individuals as well.
Democratic socialists and social democrats solve the free rider problem through general taxation and regulation.
Consider universal healthcare as the case in point for this; we absorb the cost of chronically ill people by mixing them in with the rest of the population, at a fraction of the price that the "free market" costs to attempt and fail to do the same thing.
Actually I don't think libertarians would argue that the problem is unsolved in both systems, they would argue that the problem itself is nonsensical. Most libertarians don't believe that non-excludable positive externalities are a problem, since it involves no harm to anyone and no violation of rights. They simply don't believe that because you indirectly provide a benefit to someone, you then have the right to coerce payment.
One could argue that there is an efficiency problem however - for example, take a bee keeper whoes bees benefit their neighbours. It could be argued that if there was some means to which the keeper could exclude those positive externalities, and there some level of payment at which the surrounding property owners would be indifferent between the excludable and the nonexcludable situation, there could be a Pareto-efficient gain. And since there is no reasonable way to exclude the benefits, it leads to the conclusion that the neighbours should be coerced into payment. Most libertarians reject this type of coercion prima facie.
> Libertarians can just flip it round and say how do socialists solve the free rider problem?
This is a fallacy (tu quoque/whataboutism). You're changing the subject to distract from the fundamental problem in libertarianism and implying that some other strawman is just as bad.
Without solving the fundamental problem, libertarianism will never work for anything but toy societies.
Your spidey sense is failing you, you're making assumptions you've got no ground for. I've voted for left of centre parties all my life. I was born in the NHS and I'll probably die there too.
Not really even in the same ballpark as what they did. These other labs are using AI generated content (which has already been ruled un-copyrightable) to train their models. Oh and they are paying for those tokens. So at absolute worst, they are violating the terms of service. The horror. Meanwhile these frontier AI labs pirated and scraped everything they possibly could, paying not a dime to the copyright owners, nor paying anything to the websites they DDoSed.
Data mining for AI is presumably fair use, whereas when you sign up for a Claude account, you enter into a legally binding contract that says you will not distill a model based on its outputs.
I mean I believe in protecting your company's IP, but IP and patent law is absurd these days, designed to protect investors and their fake money rather than actual inventors (who usually get no proceeds/are shafted).
They trained from the internet, so if someone trains from them it's fair game. Their clever tech should be in the mechanism with which it uses to provide an answer, not the answer itself.
Call the wambulance a company that stole all of humanities public data to train a model is mad that someone used their model to train another model.
Give me a break. Every employee of anthropic is going to have $20m or more at the IPO.
I found out today that an employee of the home care agency I own is homeless. We are trying to figure out how to help her but it's shockingly common in the industry and there are limited resources to solve the reality of working homelessness.
As a Open Source contributor who was never asked by Anthropic or OpenAI if they could use my work in their training datasets, this sounds so deliciously ironic.
This is like a Gardner complaining that you watch him as he works to learn his craft. My dude you do not have to take the job, but most people just accept it as the way the world works. If they feel like they do not want to serve the Chinese they can do that on their own, why do they need the government?
Why is it called "distillation" when it seems to be "scraping"? (as in web scraping)
When bots open the same board 1 million times per day it is web scraping to train the AI model and OK.
When someone asks 150 thousand questions it is now distilling.
On an unrleated note, 150k qieries feels like nothing?
Scrapers seem to account for 50% total internet trafic.
Do they use different methodology since it is suddenly bad when scraping happens to them?
Has anyone else noticed that Deepseek v4 running in Claude Code will try to read, list, tail as many files/logs/... as it can for even the most simple tasks?
It's so funny how LLMs, which trained on millions of books, stolen (and even if they weren't, which they were, pirated from online pirate sites like libg and annas, they didn't have consent for the VAST majority of them), and stolen code, and stolen comments, etc.
Now complain about their stuff getting "stolen"... lol.
Sounds like fair game considering Claude is built upon the theft of creative assets of the entire world and aims to eliminate white collar jobs entirely
AI is awesome tech but it’s also to some extent mass piracy. The models are trained on huge amounts of material with dubious or non existent rights.
I have a hard time being concerned about “you pirated my piracy.”
I hold the view that many of these models should not be copyrightable. Anthropic and all the others talk about “safety” but you never hear them bring up attribution of the data that trained the model or compensation of anyone for it.
Perhaps this is related to the "Mythos is too dangerous and cannot be exported" movements? It'd be a fairly effective way to justify extreme actions in combating it.
One could even wonder if they requested it, as a tactic to support their eventual IPO valuation.
Which is part of the problem of such an obviously-corrupt government: conspiracy theories are somewhat reasonable, as they keep getting validated.
So let me get this straight, a company which built its whole business on ignoring IP is all of a sudden upset that somebody is not respecting their IP?
It's hard to sympathize with Anthropic for this or the export ban, the hype over model capabilities probably fuels both things (in some ways). Training data for me, but not for thee (at any scale) doesn't seem like a tenable position. If anything, Claude's constitutional outputs should be trained on more rather than less.
Anthropic extracted millions of words of my own writing even more illicitly for they did not do so through an API provided for that purpose while paying me in the process.
"The distinction between downloading pirated copies vs. scanning physical books is fascinating — same data, different legal outcome. Copyright law really wasn't built for this era."
I like that they use “illicit” and “fraudulent” like as if model distillation is illegal and giving them money and then doing whatever they want with the output of their publicly accessible models (which Anthropic does not own) is… also illegal?
“Anthropic, red faced after unattended ice cream cone eaten by ants on park bench, once again demands government pick it as forever winner, adds ‘no take backsies’”
Oh no, the thief is mad they get "stolen" from? I've had to hard block all of anthropic's scrapers because they seemingly ignore robots.txt and every other unwritten rule about 'polite' webscraping. They were so aggressive that they made up 90+% of our traffic + pulled the website down at times. And that's not even mentioning their other immoral practices + what they are claiming here is questionable at best. Anthropic is _not_ the good guys, they do not get to be upset over this or claim any sort of moral superiority over 'China'.
Notice how Anthropic is now scapegoating Chinese models providers like Alibaba and outright accusing them of distilling their models.
Whether if it is true or not, this is part of their effort into using them as an example to scare everyone into getting congress to ban powerful models from being accessed outside of the US and also banning powerful local models from being released.
Anthropic does not care about you, and they are not your friends.
I think it’s more than that. Piecing together the perspective of a few commentators in this post - it’s plausible Anthropic is trying to shift the narrative from US vs. Rest of the world to US vs. China.
In other words, they want to sell Fable or future more powerful models to rest of the world (presumably all future models are going to be more powerful than current gen). One way they can sell this is to the government is by scapegoating China (which is their primary concern anyway).
This is working on the presumption that non-US companies form a material portion of their current revenue.
Only China really has the resources (multiple labs invested in the space), culture (Asians are generally collectively-inclined, so sharing is in their core) and political bent (there will be no diplomatic repercussions) to put up a fight.
> Only China really has the resources (multiple labs invested in the space)
That's not the point. Why is it a country thing? There are plenty of non-China startups in this space having resources at that scale. The "China" has resources is some "Western media narrative" speak. So Meta should have won a long time ago? Or xAI?
> culture (Asians are generally collectively-inclined, so sharing is in their core)
Just stereotype it? So we've gone from China -> "Asian"? Then where is your Korean or Japanese model etc? And somehow you know they're sharing.
> political bent (there will be no diplomatic repercussions) to put up a fight
More inferring from "Western media news"?
Where's the reality?
The media hyped up Gemini / Google TPU free-win last year. How did that go?
Because the China vs US geopolitical situation is a thing. Meta is a social media company, not an AI company, and they direct their focus as such. xAI just never got serious traction so now they're selling their compute. Also if a US company were caught distilling, I think Anthropic could actually take them to court, and I'd guess they don't want that kind of PR.
> Just stereotype it?
Is China not Asian? Are Asians not generally collective/cooperative, as opposed to individualistic/competitive?
The "and" that joined those 3 items is very important: it means you can't pull them apart and address them independently as they each contribute to the context. I'm not too sure about Korea, but in a way Japan is a US colony in all but name. Both are very much politically intertwined with the West (along with RoC/Taiwan), which means nothing major that may be against US interest happens.
The reality is that China and the US are essentially in a trade war, where the latter is trying its best to keep the former in the Dark Ages, because "national security", but the former is refusing to take it lying down and continues to make progress regardless[0], because they have the resources and will.
> Because the China vs US geopolitical situation is a thing.
By the media? It's easy to point fingers at a blackhole.
> Meta is a social media company, not an AI company,
Alibaba (the discussion here) is not an AI company too (by your definition).
> Also if a US company were caught distilling, I think Anthropic could actually take them to court, and I'd guess they don't want that kind of PR.
Meta has been to congress. Microsoft, Google etc have been in lots of court cases and continue to do so. Do you really think that is what stops them?
> Is China not Asian? Are Asians not generally collective/cooperative, as opposed to individualistic/competitive?
This is exactly the "media" view you get. It's just stereotypes and generalization.
And yes, that is wrong by the way. Evident in real data. "China" as a whole wins market share in many areas but no 1 company has as much of a monopoly as US companies do. Why? There's so much competition that it is scary. So are you sure they don't compete?
> but in a way Japan is a US colony in all but name
Again, I almost give up seeing this. Clearly, not. If a whole country, the world's top 5 in GDP is only that to you something is wrong with what you're seeing - not with the country.
> Both are very much politically intertwined with the West (along with RoC/Taiwan), which means nothing major that may be against US interest happens
On the table? You do know that China is a top trading partner with all of these on your list. Despite whatever spat you might see in the media.
> The reality is that China and the US are essentially in a trade war
No. That's what the US government wants you to believe. It was even documented that in his 1st term, Trump, wanting a grand policy asked Krushner, whom then suggested China (pretty randomly) and so they went with it. Trump has now done less "China" related things lately due to all the backlash that you'd think he has moved on and found new toys.
Until very recently, the export ban GPUs had such a loophole that Chinese companies were able to use subsidiaries outside of China to buy and train that the whole thing was meaningless.
i.e. conclusion: stop getting brainwashed by media articles. It's all a show to get someone like you riled up.
I would say Antrophic and others illicitly extracted free internet content and put it behind a paywall, giving zero compensation to those that made their whole business possible in the first place. So smallest violin player busy here trying to make me care if it happens to them.
fucking lol. it is always funny when companies use opensource and other free for non commercial use - and plain old piracy - and then cry about the same practices.
> Meanwhile, on June 12, two days after Anthropic sent the letter, the Commerce Department imposed controversial restrictions on Anthropic's latest Mythos and Fable AI models because officials feared they could be deployed by military intelligence users in China and other countries of concern.
So that was the real reason for the Fable restriction? Because Anthropic wrote a letter to the US government saying that China was distilling Fable?
In an other news, a terrorist organization practicing torture at daily level just released a public denunciation of the evil forces they are fighting against, guided by their holy mission of making progress in social morality for all of us.
Gosh, overusing accounts running up unplanned-for expenses?
Kinda reminds me of...overusage charges and inflated expenses clients have had to deal with because Anthropic, OpenAI, Grok, etc have been "illicitly extracting" everything they can grab from said websites, as fast as they can. In what amounts to a DDOS, frankly.
It's a bit disappointing when people see this as a schadenfreude moment because it's clearly not safe nor a good precedent for potentially dangerous AI models to trivially fall into the hands of malicious actors.
Oh, c'mon. If Alibaba wanted, it can have the entire Claude/Mythos source code and data by next week. All you need is enough bribe to a developer that has access to the repository. Humans are always the weakest link in anything.
This article is absurd for an outlet who published an article that's meant to be news not editorial. Reuters was once a news wire and is still considered that. The first two paragraphs refer to "attack" and "strike" against Anthropic. This is sensational nonsense, not news.
There was no strike, or attack. Block the accounts. Why is this news, and why are they pandering to the people who just banned the new model they burned at least $10 billion training?
The closer you look at this AI stuff the more absurd it is. I assume the strat is to keep the bubble floating until post-2028, then drop the bomb on the Dem who wins. Just like with the covid inflation + economic rigging Trump did in 2016-2020.
Suppose Anthropic trained only on data they paid to create, and not the internet or stolen textbooks.
It would still be extremely difficult to muster any sympathy for an organization whose MO is to go public not to honestly raise capital to fund growth and development, but rather to dishonestly leave someone else holding the bag, in some cases involuntarily as their retirement funds are passively invested.
And even supposing they were honest and didn't have an IPO, it would still be extraordinarily difficult to care about their misfortune, because "consolidating all thought-work into the hands of those few who can afford frontier models and datacenters and power plants" is also a special kind of misanthropy.
And even if that were not the case, they're filthy rich already, so who gives a shit if the Chinese companies prevent them from becoming quadrillionaires? :)
Ask claude it's name in chinese and it thinks its Qwen (opus) or Deepseek (sonnet). Anthropic are just as guilty as everyone else training AI, today, maybe more so. Every lab borrows from every other. It only takes a few hundred samples to figure out the pattern; look at glm-5.2 reasoning using the caveman tongue of gpt-5.5. Stopping this would require some draconian surveillance.
That's not how it works though. When you prepare the conversations for distillation, it's the most trivial and obvious first step to replace "Qwen" with "Claude" and vice versa. I doubt they'd simply forget to do it.
A model may misidentify itself due to the surrounding context. When a model is about to answer "I'm ...", what follows is a sorted list of probabilities for what the next token should be. In most models it's usually a list of popular model names: say, in the list, first comes Claude, then Qwen, then ChatGPT etc. Usually the "Claude" token would be the most probable token, say 70%. But if the surrounding context is in Chinese, the embeddings for "something to do with China" may nudge the combined embedding of the output token towards the "Qwen" embedding more ("China+Claude=Qwen" in the embedding space). Say, the probability for "Qwen" now becomes 60% instead of 10%.
If we also use high temperature for more "creativity", the token sampler now may choose "Qwen". It's not the most probable token still, but it was chosen because selecting the 2nd most probable token once in a while usually allows a model to explore unexpected "creative" paths, and 60% probability is good enough compared to 70%. It's basically a hallucination.
I once made an experiment: if I ban the word "Qwen" in the inference engine entirely, and ask Qwen "which model are you?", it happily starts announcing it's Claude 100% time, simply because "Claude" is the next most probable token after "Qwen" in this context.
There's two basic kinds of distillation: 1) the massive [and dumb] method where you ask a question and use the answer as reinforcement (Black Box), and 2) more targeted distillation where you use one model to directly inform/train/guide another model (RLAIF).
The latter is basically fine-tuning the model with direction from another model. Thousands of businesses do this every day to fine-tune. This is almost certainly what the Chinese labs are doing, since it has a much better effect on the end result than just getting simple answers to simple questions.
These complaints of distillation are inflating the problem to make it sound worse than it is, because they want the USG to block/ban Chinese model providers as protectionism. They have already called for more export controls on chips (which is funny because DeepSeek v4 was designed to run on Huawei chips and now the other Chinese providers are following suit). But they can't come right out and say that, so their claim is that they're asking for more export controls because distilled models might not be as safe as their own. But if you show them a jailbreak of their model that bypasses their safety, they'll tell you that any model can eventually be jailbroken so don't worry about safety.
> These complaints of distillation are inflating the problem to make it sound worse than it is
Unfortunately, the Reuters piece itself is complicit in this dramatization. The lede paragraph parrots Anthropic's talking point that distillation is an "attack", without using quotes that would alert the reader that this framing is a corporate talking point. Distillation is NOT an attack.
Agreed! I had to do a double take and check the URL. I thought I am reading a press release rather than actual reporting.
That's exactly what they pay the publicist for.
https://news.ycombinator.com/item?id=13155538
ironically, I think this is why the jobs apocolypse is overblown, Ai is only good at a thing if the people using it are also good at that thing, and people are attributing Ai as superhuman at things they do not know themselves
AI doesn't have to be able to do your job to convince your boss that it does
Same thing nowadays :^)
It always was.
> Distillation is NOT an attack.
From the article -
> 28.8 million exchanges with Claude through almost 25,000 fraudulent accounts
wouldn't that be considered an attack? Not sure what I'm missing here.
An attack against what? The sanctity of "their IP" that is itself the result of a massive copyright violation campaign?
Has it been proved in a court of law that it is a copyright violation?
In some cases if the model regurgitates the original material then that is clearly copyright violation, but if the model "learns" from the source material just like a human brain would then that's not a copyright violation.
No, what was proved in court was that they downloaded and trained on millions of pirated books. The court said their use of books is fair use, but stealing them isn't.
I think we're going to see cases that find distillation is also fair use. You're using the competing model like a book. You pay for it, you use it (read it), it informs your model, but you aren't repeating/reselling what the model told you verbatim. Foreign labs may still run afoul of competing labs' Terms of Service, and they may also pay a settlement (or not, it's a different jurisdiction after all), but the damage is already done. Distillation will become uncontroversial when done legally.
Are LLMs even copyrightable? If not, no need to speculate fair use.
Then distillation isn't a violation either by extension.
I would agree, if they are inspecting static output of American AI models without using their compute resources.
Scraping the internet for training is also using compute resources.
Aren't they buying the use of these resources just like any other customer?
> Has it been proved in a court of law that it is a copyright violation?
God I'm so tired of this.
The billion dollar companies have the ability to hire an army of lawyers to DDOS the legal system. They at most pay a slap-on-the-wrist fine as the cost of doing business.
Ddos is a great framing of this :)
I'm extremely pro free markets etc, but the uncomfortable truth is anthropic stole the work thousands of authors for profit. I think it will one my favourite things in life: programming books.
it's a 'too big to fail' model. Because they have a big swinging dick all the copyright and other restrictions they violated would nuke them from orbit so we can't actually hold them to account for it .... for some fucking reason.
even if you disregard training costs, pure inference costs are a problem same reason other api have rate limit. this is an attack to bypass the rate limit.
Be careful to properly identify the bad behavior. A customer who buys a product for less money than it cost to produce has not necessarily done anything wrong. They just took advantage of a loss leader. That's on the seller.
Did you notice that when Valve was displeased about scalpers, Valve changed Valve's behavior?
It doesn't seem reasonable to complain that a customer of your AI service received that service for less money than it cost you to provide that service. I don't think that is the complaint here at all. If that was the issue, they could just raise their price.
As most everybody seems to notice, this is just a reenactment of what was once written for comedic effect: "You're trying to kidnap what I have rightfully stolen!"
Perhaps an arrangement can be reached.
https://clip.cafe/the-princess-bride-1987/youre-trying-kidna...
Still calling it an "attack" feels like a stretch.
They literally had to pay for that "attack", no matter how many accounts they used.
Google was killing many websites for decades with their crawlers. Most large websites decided to create dedicated infrastructure for their traffic alone. Somehow they didn't participate in that cost and were not called the attackers.
> and were not called the attackers.
This is the mental mental leaps I'm struggling with here. Did you not live through that era where they were explicitly and repeatedly called out as 'attacks'? They were generally tolerated/hardenee around as they provided value-in-discoverability.
Just to ensure you don't gaslight yourself - I did live through that era and I worked on and supported a niche community (a MUD) where we did a lot of work encouraging marketing and discoverability through MUD forums as well as making sure our page was accurately and minimally keyword tagged and highly available for indexers.
In the time since that era search engines have transformed into platforms themselves that do engage in more parasitic behavior but it's important not to assume that the way it is now is how it always was - that's a rather defeatist path to walk down where you ignore awareness of the fact that there can be a highly profitable non-enshittified search engine that supports, rather than destroys, the ecosystem it benefits from.
It was better and, if we're diligent, it can be better again.
> [Google] were not called the attackers.
They should be. But as the saying goes, one website/company dying is a "tragedy," lots of them dying at the hands of one company is a statistic of corporate growth. Or something like that.
And then of course when the tables turn on a company and they're the ones getting bombarded, they cry foul. Keep in mind Anthropic did many similar things that you mentioned Google did.
I think the term "attack" here is appropriate but not in the way Anthropic is framing it. Alibaba is clearly violating terms to extract data, so that's definitely not above board. But it's not like a DDOS attack where Alibaba is trying to attack Anthropics servers. Alibaba is simply doing exactly what Anthropic did to the rest of the internet, just targeting Anthropic and paying them to do so.
It's merely a ToS violation.
My terms of service are that you are not allowed to breath oxygen.
I am getting a bit tired of companies being able to have user hostile, anticompetitive, monopolistic terms of service. The freedom we give them comes at the cost of the freedom as consumers to have free markets because they lock them up
Exactly, calling it “illicit” is funny. Your ToS isn’t law.
Illicit means maybe against the law but definitely against the rules, for example an illicit affair. The word for against the law is illegal, from Latin, or unlawful, from Germanic. I guess the Germanic cousin of "illicit" would be "forbidden."
Extramarital affairs are against the law in many countries and 17 US states. “Illicit affair” is potentially a holdover from when it was illegal more places, not just a conflating of against the rules with illegality.
https://en.wikipedia.org/wiki/Adultery_laws
Attack or customer
That's violating TOS, spamming, possibly a DDOS, but the distillation in and of itself is not an attack it's just using the model.
Like the difference between scraping a site with one or two active connections vs thousands. It's not the scraping that is an attack, it is how they are going about it
> That's violating TOS, spamming, possibly a DDOS
As in distributed distillation of service?
Just sending a request to a service does not constitute an "attack". It seems that what Anthropic mean by "fraudulent account" is probably just one violating their terms of service - misuse of a subscription account, and/or the presumed nature of what the user was trying to do.
I guess Anthropoic would regard any developer using their subscription plan with OpenCode to be operating a "fraudulent account", maybe an "attacker" too. Now we know how they think of anyone using Claude to develop software competing with Anthropic. Only an "attacker" would want to vibe code their own harness, or god forbid want to learn how to build/train an LLM.
Of course Anthropic's wording is intended to be deliberately provocative, since they are trying to manipulate the US government into shutting down the Chinese competition.
Is an attempt to copy all or parts of a model an attack, when models have very questionable copyright status? Maybe? I don't think most people have much sympathy here though.
Let’s not forget that by the same logic, Anthropic et al are “attacking” copyright holders all around the world by scraping their data unauthorized for training.
Pot calling kettle black.
Not only that, daily flooding websites with almost infinite amounts of request for ”web searches”. DDoS-by-VC money.
i mean, i got 5 replies in a minute of asking, and none deny it's an "attack", they simply say "good". HN should be better discourse.
https://en.wikipedia.org/wiki/Ken_McElroy
Distillation done via bulk automated activity of fraudulent accounts, in violation of a terms-of-service, can reasonably be called a "an attack" – specifically a "distillation attack" – even though distillation itself isn't necessarily an "attack".
This is similar to how compromising an account through bulk automated trials of many passwords is reasonably called "an attack" – specifically a "dictionary attack" – even though using a dictionary is not itself an "attack".
You shouldn't need to smuggle your sympathies (for the tactic or perpetrators) or antipathies (for the target) into peculiar judgy language prescriptivism against common, understood usages.… that then label Reuters "complicit" for simply reporting Anthropic's claims accurately. That's what Reuters is supposed to do, in a story about a letter Anthropic wrote!
Labeling it as an attack is smuggling sympathies. It is not common; there are only a small number of people who even discuss the concept. A company buying a product with the intent to reverse engineer or copy its features is likewise not an attack; it's just normal competition that benefits society.
The standard of neutrality that people here pretend to require from news organizations is not even remotely realistic.
It was a timely story from Reuters. They do fast news feeds, like APnews. Could it have been better or more accurate? Sure, they could have gone into why distillation may or may not be seen as "an attack". But then it would have been a more involved story, defeating the purpose of a news feed.
The Reuters piece was "good enough". Some other place like the NYTimes or WSJ can follow up with more detailed investigative coverage if it's a worthwhile story.
I don’t want or need fast and “good enough” news and i’m gonna try and make a case that you don’t either.
Until very recently, all of modern civilization was built by people who got their news at most once a day. Reputable bureaus like Reuters took that day to get it right.
I’m not the national security advisor, so I don’t need a push notification that there was an earthquake in Nepal, or a bullshit rush-job briefing on Chinese AI distillation tactics.
The fast part isn’t for your benefit, primarily, and news media would love to go slower and have more time if they could, and still survive. The race to break news first - in order to be the one to tell their audience something “new”, something they hadn’t heard elsewhere - is real and it has been around for all of modern civilization, for hundreds if not thousands of years. A one day turnaround was a thing purely due to daily newspaper print runs being the fastest distribution, it wasn’t because it was long enough to get it right. The reason they had a day is because the competition couldn’t get something out faster than that. Then for a while there were twice daily print runs to be more competitive. Then the internet came along, and now the only way for a site to get attention and be talked about on Hacker News is to report it before any other sites do.
There are some news media that do go slower and take their time, but I think they’re struggling to stay alive. Reuters is still reputable, but they no longer necessarily take a day. The big question is how do we get humanity to prefer slow & correct over fast, and it is even possible? When you hear about an earthquake in Venezuela, how do we stop people from Googling it immediately, and get them to wait for the best most correct story rather than reading whatever’s available now? In the case of natural disasters, I don’t think it’s possible anymore, no matter what case you make. I’m not sure it’s possible with stories like AI distillation either, even if you can absolutely cement the case for slow news. The fact that it’s async/internet now and that first still counts means we (you and I) are still going to give traffic and attention to sites that have the first information on a breaking topic, statistically, despite having a preference for correctness over speed. The one thing we can do is vote with our dollars by subscribing to whatever news media that does a better job than others.
Good enough slop to serve the masses. Doesn't need to be truthful because its fast? Why even both to write anything?
Yes. It was good enough to communicate that news item.
Did Alibaba perform "an attack" or were they taking advantage of resources and going beyond Anthropic's terms of service? Didn't Anthropic do the same kinds of things when building their models?
These are all interesting questions, but they don't have to be addressed in full by a news blurb about a letter Anthropic wrote to some senators.
Money. More eyeballs on it means more ad impressions. Same thing with 24 hour news channels.
Distillation may not be an attack, but it is a ToS violation and could be seen as IP theft.
Any reasonable company would be pissed if a competitor, especially at Ali Baba's size, leveraged that company's R&D to compete. It is in this sense, a corporate attack.
If you want to roll your eyes at distillation concerns, you might need to excuse Anthropic for originally using pirated material to train their models.
More the opposite - companies who stole IP for their own benefit have no leg to stand on when others do it back. Personally I couldnt care less if Chinese labs rip off Anthropic. Its what America would do if they wanted to, for whatever reason (they probably do it right back secretly anyway).
Reuters is probably the most rigorous news agency in the world.
> it said was the largest known attack
> Anthropic said in the letter it was supportive of the U.S. government's efforts to combat the attacks
both times the word "attack" appears it's clearly stated that the word was used by the company, it's a direct company quote.
actually putting it into quotes would be editorializing
> Unfortunately, the Reuters piece itself is complicit in this dramatization
how would you feel if somebody quoting you would turn your word dramatization into "dramatization" because they don't agree with your assesment
> how would you feel if somebody quoting you would turn your word dramatization into "dramatization" because they don't agree with your assesment
This is exactly what news agency should be doing though. When the dude showed up to Comet Pizza to look for Hillary Clinton or whatever, do you figure they should've printed "Local hero saves children from predatory cabal"?
I want them to report the facts, not their opinions.
Reporting that corporate called it attacks is good. I do prefer direct quotes.
However, when they quote one word, the journalists are inserting their own opinion about it. I want to make my own opinions based on the facts. I don't need the reporter to draw the conclusions for me.
Well, let’s say you put the picture of some political figure, and put in highly contrasted red, bold large catchy font, "TERRORIST THAT KILLED MILLION PEOPLE", then below that in barely visible contrast, in tiny discrete letters, "is what this person probably will claim to be against".
This whole sentence technically will be correct, 100% guarantee, whatever this person actually even said or think.
From a propaganda point of view, framing the elements of language is even more important than what the statements actually states to be true or possibly true.
nice slippery slope you manufactured there - what if Reuters becomes Daily Mail
what framing are you talking about? they are literally quoting a company.
please explain what Reuters should have done here. Should they have added in parentheses: (editor note: we don't agree with Anthropic calling this an "attack")
Is that what you want? News outlets giving their opinion and moral judgement on company quotes? I mean, Fox News/CNN do have a large following, so there is clearly a market for that.
> please explain what Reuters should have done here
This is very straightforward: use direct quotes or use neutral language. The article describes the alleged incident as both an “attack” and a “strike” in the first two paragraphs. And neither is within verbatim quoted text.
Reuters, however highly you may regard them, simply adopted Anthropic’s framing uncritically in this instance.
You are confusing stylistic choice with framing.
A lot of times Reuters paraphrases instead of "quoting quotes".
> "uncritically"
You are mistaking Reuters with CNN or FoxNews. If you want "critical" reporting you should read some bloggers instead of news agencies.
If you’re going to call out their use of slippery slope as a fallacy then it should be pointed out that your original argument was framed on an appeal to authority of Reuters as a leading news agency.
Both are logically unsound.
Anthropic raped everyone without asking and stole their labor to build their career-commoditizing tech.
Distillation is Robin Hooding it back so that one trillion dollar company doesn't reap all the benefits of their automation of the workforce.
Distillation is Prometheus bringing fire from the gods to give to ordinary humans. Something we all own anyway, but that was kept from us.
Distillation is freedom.
Everyone should be pro-distillation. We should all work together to distill every proprietary model.
Anthropic stole. OpenAI stole. Google stole. ElevenLabs stole. Suno stole.
We should be able to get it all back.
And a number of Qwen variants are available to self host. Do Anthropic have any like that?
I'm more excited by open weights models you can't self host and need to spin up on H200s (RunPod or bare metal). This is where the real power lies and is where the open source world will trend.
It's far cheaper to spin up an H200 hourly or to simply consume a managed version of an open weights model than it is to use a proprietary hyperscaler API. And you own the model itself and can fine tune, tweak, lobotomize, etc.
The stuff you can run on your own RTX cards is neat, but it's rather hobbyist. The real power is in the cloud. Renting cloud hardware is fine, because the core problem is ownership of the weights, not the server rack or ISP fiber lines. Those are already commodity.
Big businesses will eventually run open weights models in the cloud, and it'll be a rather large part of the future AI economy.
Eaaaaasy now, the Chinese labs aren't freedom fighters on behalf the common man. They're not non-profits, they're not neutral transnational organizations only dedicated to open source efforts.
They're Chinese companies offering open source models now as loss leaders to keep themselves in the game because they know virtually nobody, especially in the corporate world, would contract with them and give them access to their data. They might as well just send a Dropbox link of all their sensitive data directly to their Chinese competitors, same end effect.
They're also doing it as the digital equivalent of what they've done in other industrial sectors for decades. Undercut and flood the market and once you've killed or severely hindered your competition, then you have the market cornered. The moment they can afford to these open source releases will stop.
Then the world will be stuck, just the way the world is largely stuck on rare earths. Instead of being able to regulate the leading companies from DC and Brussels, they'll be stuck watching Beijing call the shots.
That world would likely always have guys like Mistral and Trinity, but it's an open question if they'll ever catch up to the frontier.
And then Beijing will enjoy access to the data (ask any multinational operating in China for more than 2 seconds how useful contracts and Chinas legal system is for protecting IP), and these companies will roll in the money, and the Chinese supply chain will grow up behind the labs.
So, let's not pretend they've got the moral high ground. No. That boot just isn't on your neck yet. They're playing the long game -- and they're good at it.
I think most of us know why they're doing it. We are just very pleased with it regardless.
1. I get great products for nearly free 2. Anthropic/openai/etc will hopefully be destroyed since they stole everyone's work and are trying to capitalize on pure theft.
Win-win. The why of it is not really that relevant.
>We are just very pleased with it regardless
You don't trust the multi-billion dollar behemoth, but you trust the militarized multi-trillion dollar behemoth to play 'robin hood'?
i can't get my brain around the mental loops here.
If you don't think Anthropic and OpenAI are multi-trillion dollar militarized behemoths you need to catch up on some news.
Both are planning $trillion+ IPOs this year. OpenAI is collaborating with the Department of War, and Anthropic is under intense pressure to do the same and their top model is being held hostage right now. This week, the Department of War wrote a statement that xAI should not be held accountable for environmental laws because Grok is a vital weapon system of the US and was used to fire over 2000 missiles at Iran. The pentagon's statement mentions there are 3-4 such models so you may be able to guess which they are.
I don't get it? I use the open weights deepseek on opencode Go hosted in the us/etc.
What are the mental loops here?
I would genuinely like to know if I'm missing something.
> You don't trust the multi-billion dollar behemoth, but you trust the militarized multi-trillion dollar behemoth to play 'robin hood'?
Nobody's trusting anyone, we're just enjoying the benefits of true competition much like the working middle class gained benefits between the ideological competition of the Cold War.
The Chinese companies don't have to be open weights, and it's not all about competing with the west. For example, most of Ziphu's (GLM) business in China is supporting private on-prem instances rather than selling API access. They make money by selling support services - much like RedHat's busines model.
It doesn't matter why Chinese firms are stealing models and open sourcing them. The fact that they are doing it is a very, very good thing for basically everyone other than the people who paid to build the original models, but I've got no sympathy for them considering they stole all the content to train them in the first place. This is some kind of beautiful irony.
> it is a very, very good thing for basically everyone other than the people who paid to build the original models
It's not a good thing if you think there's more discovery and progress to be made, rather than cannibalising a fully mature field with cheaper alternatives. Drowning R&D early is not good for everyone.
What does further progress get us? Mass unemployment? Extinction? Pick your dark future science fiction?
The happy ending where we're all living in a garden of eden cared for by benevolent AI is hardly worth considering when you look at the cast of characters who are in charge of the world right now.
Is leveraging an enormous capital advantage to strip-mine the Internet and sell it back to us cannibalism or not? Confused on this point. I think they are exploiting a loophole in copyright law (and kind of redefining the meaning of "derivative work" in my opinion, but hey I'm not a lawyer) that collectively we tolerate because the end result is so useful
I think that's a slightly different topic, but: a) strip-mining the internet is definitely the most misleading way to think about it. Strip mining means aggressively removing something to the area's detriment, and nothing has been removed. If all AI is turned off today the internet has not lost all of its natural resources, and silly phrases like that fuel inappropriate emotions and consequent conclusions and b) the internet is not being sold back to us - that is also a highly misleading phrase, if not an outright lie. The internet is still there and we can use it. No one is selling back to us what we already had. AI is not the internet cordoned off and resold.
I don’t think many outside the US are actively hoping to be governed by Sam, Dario and Elon.
The "why" always matters in everything in life.
Can you please tell my, as someone who is neither Chinese nor American, "why" I should care if a Chinese company stole from another American company (that in turn stole from everyone) to give me a cheaper service that fits my use case?
> to give me a cheaper service that fits my use case?
Because they aren't giving you a cheaper service that fits your use case.
Best Case scenario, it's a trillion-dollar behemoth stealing from a billion-dollar behemoth so they can add their own explicit restrictions/weights on top to influence the masses.
There is no 'robin hood' here, any perceived value you get is clearly and explicitly tainted. "I don't care if it doesn't show me non-party-line results - It makes me a cheap UI !". Ethics/morals be damned.
> There is no 'robin hood' here, any perceived value you get is clearly and explicitly tainted. "I don't care if it doesn't show me non-party-line results - It makes me a cheap UI !". Ethics/morals be damned.
I can't tell if you are talking about Anthropic or Alibaba here.
and honestly that's my entire point. There is no Good Guy here.
In a world which already has the likes of Anthropic and OpenAI, having Chinese labs be a counter balance is decidedly better than the hypothetical where American companies had a global monopoly on LLMs.
If your argument is that all present LLM offerings are unethical then that is something I am sypmathetic to. That said, I am also unable to offer a conceivable roadmap to undoing the opening of the LLM Pandora's box so I tend not ground my arguments in anti-LLM advocacy; that would be very 2023 of me.
The whole AI industry was built upon stealing IP.
The extreme of this is to make IP laws irrelevant and that everything should be in the public domain.
Which maybe is not a bad outcome for humanity as a collective after all.
The main problem is how they accessed the IP, but then using it to train a model is fair use. But yeah, IP theft doesn't exist because nothing is stolen really: Hollywood studios still have their movies.
Um, yeah. They stole the IP and then they stored the pirated IP. It was literally stolen and stored on their servers. That proves that IP theft exists. It's not complicated.
I don't think that's true. Sometimes the 'why' is lost in time as no one's around to tell it, so we end up with a "if a tree falls in the woods and no one's around to hear it, does it make a sound?" scenario. It doesn't really matter. The thing now exists without a 'why.'
you dont get it - usa is the goliath in all scenarios online. these are us based companies. most of the world would like to see them and the us fail.
They want to create a monopoly and destroy every competitor, before they got a chance to rival them.
Why can't OSS software rival closed source software? It should be an open market, at least "somewhat", what's happening for real? EU providers will also get banned, if they reach or exceed US model capabilties?
Closed source providers can close your account at a whim like and destroy your business and then use the data you supplied them to create a competitor (Meta, Google, OpenAI, Anthrophic).
Well Zai's GLM 5.2 legitimately is a frontier-level model, though not quite parallel with Opus or Fable. Unfortunately, its too damn big to run locally for most people. Thats the bottleneck right now; the open-weight models exist but something capable of competing with the frontier models just can't run on anything normal yet.
>They want to create a monopoly and destroy every competitor, before they got a chance to rival them.
VC/Startup playbook 101.
also why cant i have my own airport, too big to fit in my backyard... you guys lol.
https://research.nvidia.com/labs/lpr/slm-agents/ - Distillation data is a natural byproduct of using these models. There's no effective defence against it. Anthropic is degrading thinking blocks to summaries to slow it down and hide model internals, but in the end, the math says you're SOL and it works on MNC/Large Corporate scale well enough that the moment cost becomes a priority, you're left without the lock in you need to keep customers paying.
Byproduct? It’s essentially the only part of an LLM that is useful, because it’s the WHOLE product!
It’s the same reason why DRM for audio and video is a non sequitur - if you want a person to see or hear audio or video, eventually at the end of the chain, it’s going to be converted to audio for the ear and light for the eyes - that’s why you attach your tap.
Without a model generating tokens, what’s the point. So if Anthropic somehow disable quality token generation, what’s the point!
That's why the harness is moving server-side: because generating tokens is not the actual point of the model, not for the users. Especially with tool calling giving us agents that can act, most of the tokens generated are not, themselves, critical to the end users. Specifically, a lot of tokens goes into orchestrating actual tool calls, and then most "thinking tokens" are only relevant to users only in so far as they help users keep track of and verify what the LLM is doing. So all those tokens can be hidden or replaced by partial summaries, and all of that can happen server-side, and then there's very little to distill from.
I haven't heard of this happening, do you have links any explainers on this?
Heck, one of my favorite fine tuned copies of Qwen uses Opus 4.6 Reasoning distilled. I'm not sure where the maintainer is based out of, but me in the states, if I had the hardware to do similar things I would. Its like you say, basically everyone is doing it. It kind of makes sense to me too given that you can have roughly similar data, but your reasoning logic is what the real secret sauce is in my eyes. It doesn't matter if you know everything in the world, if you don't know how to reason with that information.
https://huggingface.co/Jackrong/Qwen3.5-27B-Claude-4.6-Opus-...
It's about training data and using Claude to compare 2 outputs and have it indicate the better one. This gives you higher quality training data that you can use to train a fresh set of weights. Weights don't get adjusted on-the-fly, instead the dataset for training is improved and then you train a'fresh. And it's hard to detect because you're just asking the model which of these outputs for a given prompt is better? Or something along those lines.
Stupid question: I was under the impression that these models were trained on PB of data. Surely the amount of questions/response they can extract from querying a bigger model (Claude) is fairly modest. How is it not a drop vs the training dataset?
It's not about how big your dataset is - it's about how you use it.
I jest, but I'm also completely serious. 1T tokens from Claude can teach a model something 1T tokens scraped from the open web can't. Things like "how an LLM can problem solve effectively", or "how an LLM should use tools", or "how to construct reasoning chains", or "when to double check", or "what innate capabilities an LLM can or can't rely on".
Those are valuable things that Anthropic's own team spent a lot of effort post-training into Claude. Distillation allows them to be extracted and transferred to an otherwise unremarkable base model.
Unremarkable base model will remain an unremarkable fine-tuned model that memorised a couple thousand of input-output pairings.
Ha ha, as if.
Base models have a lot of capabilities - arranged in all the wrong ways for high performance reasoning and problem-solving. The power of fine tuning on "a couple thousand of input-output pairings" is that it can fix some of that. If your pairings are very well chosen, that is.
If that were the case, Anthropic wouldn't be throwing a fit over distillation "attacks".
Why? They often don't make sense. They send DMCA takedowns over materials they can't even copyright, for example. They fessed up to creating shadow libraries that they didn't even use in their training corpus, resulting in the largest copyright settlement ever. Your reasoning is flawed.
Yes, neural networks are famously poor at generalising.
They are poor at generalising from a small number of examples; this is why the real generalisation power is achieved in pre-training.
Can you back up this with hard data and evidence?
Most research converges to the idea that RL on synthetic data makes models worse, not better.
If what you claim was anywhere near that relevant, than we would've long achieved singularity by simply feeding increasingly better output to the training of the next model in a loop. Yet this doesn't work.
25 million turns on Claude output is a small amount, yet an expensive one (we talking hundreds of $ millions) that is better spent on compute.
There's no evidence such a process works, but I'd like to know more if I'm wrong.
> Most research converges to the idea that RL on synthetic data makes models worse, not better.
You are missing a mountain of nuance by generalizing the existence of a hole there.
Back up what? That distilling from a more capable model into a less capable model pulls the student model's capabilities up? What. Why the fuck is this even a question.
Look up literally any distillation works. Because this is just distillation but on one-hot token chains instead of richer logit KL proxies.
And no, I'm not claiming than you can "close the loop" and get RSI on the cheap just by distilling forever. I'm claiming that distillation is a very cheap way to bring the performance of a less capable model closer to that of a more capable model. It doesn't give you "a more capable model" out of thin air.
Which is why Chinese labs rely on Anthropic to provide that "more capable model" to them. They take the capabilities Anthropic trained for the hard way, and train for them the easy way.
It's a "fast follower"/"improved capability density" trick, not a "singularity tomorrow" trick. There are a few "distillation pump" tricks that get closer to what you have in mind, but they're still more about "extract more training signal out of the same set of data" than about "unbounded RSI".
so the way llms work in the first place. training on original research that was acquired the hard way.
Okay, you have no data nor evidence nor a paper backing this claim, it's just speculation.
You want to sell me the idea they are spending hundreds of millions to get unchecked Q/As with reasoning redacted and without checks on the output quality to do what exactly?
Have a shallow pointless bunch of expensive data to get slightly better RL? It's expensive and pointless.
Data has shown again and again that synthetic input/output does not benefit models in RL, it may even make the output worse.
Also, you have a giant bias.
The chinese are the only ones releasing models and research papers in the open from which American labs benefit 24/7 (DeepSeek has been copied by all US providers).
And you want to sell me this ridiculous idea of the giant return of spending hundreds of millions on unredacted pointless QAs?
What the fuck. Are you a literal, honest to god distillation denier? Straight up "wake up sheeple, model distillation isn't real"?
I've seen plenty of things in the dumpsters of AI discourse, but this got to be among the most baffling.
Yes, there are "giant returns" on distilling from a more capable model into a less capable model. And even more so when the more capable model was trained for something you want and lack. Like: better coding performance.
Someone like OpenAI had to RLVR for it the hard way (and if you think "distillation is expensive", wait till you hear how many bits per rollout hardcore RLVR gets you), but you get to peek into the results of their work and copy them for yourself.
Also, Anthropic didn't redact model reasoning until Mythos. OpenAI started with o1, but Claude had reasoning chains accessible for a long time. Which is why Anthropic was more targeted than OpenAI.
So we're meant to believe that only US companies have the intelligence and/or access to manpower to generate their own reasoning data? Does China have a population deficit? Maybe China has too high wages to pay people to generate reasoning data?
The US companies bootstrapped themselves from one model generation to the next, partly by using the previous generation to generate synthetic data, etc, and partly by paying people to hand generate training data for them. Why do you apparently assume that the Chinese can't do the exact same thing?!
Surely "coding performance" is by far the easiest thing to generate your own RLVF data for, since it has trivial verifiable rewards - does the code compile and do what you want.
RLVR is the poster child for model distillation. Because: have you considered just how many tokens does a model have to generate before you can check "does the code compile and do what you want"?
You generate 90000 tokens worth of rollout and get a verifiable reward once. RLVR is fucking expensive! It's worth it, because it often unlocks capability advances that other things don't. But it's still fucking expensive. RLVR eats compute like nothing else.
So, if someone used a lot of RLVR to improve a capability? Just distill from that "someone" and get a similar improvement for a fraction of the price! Then you can do your own RLVR from THAT cheap starting point, if you want to.
"Human domain experts" is a similar niche. Let's say hypothetical "EconomicsAI" hired some $200 per hour human economists to make training data for their "EconGPT" AI. What's cheaper - hiring your own $200 per hour economists, or using a bunch of "$10 per 1M tokens" outputs of EconGPT to bring your own model in line with what EconGPT can do?
Even synthetics can be expensive, because while synthetic tokens themselves are relatively cheap, the applied AI knowledge one needs to make high quality synthetics that improve task performance and don't backfire on you isn't. Again: distillation bypasses a lot of that - by cribbing from the outputs of a model someone has already done that for. Allowing you to get more oomph for cheaper, and spend your R&D effort elsewhere.
Your training cost argument makes no sense. It doesn't matter whether you are using human written code or someone else's LLM generated code to train on - you are going to be RL training on it, so your RL training cost is the same.
There is a data cost argument, especially if you are paying for human generated data, although I'm not sure how applicable that is to coding.
If your claim is so solid, you'll have no problem pointing out data or evidence.
DeepSeek R1 was a famous case - not only it briefly beat then-SOTA on the cheap, it was also released with distilled versions that preserved bulk of the improvements but could be run on higher-end consumer hardware.
And of course Gemma models are said to be distillations of Gemini.
The distillation you're talking about is about cutting the number of weights, it has nothing to do with extracting QAs from another model.
There are multiple stages of training, and the data/compute mix at each are quite different and produce different "layers" of intelligence.
The pretraining stage is the first stage which consists of "next token prediction" on the entire internet, PB of tokens, etc. This is what most people think of when they think of training LLMs, however it produces a "base model" which is not really "intelligent", but rather much like a blurry JPEG of all human language and knowledge. You cannot really talk to such a model; it will simply complete your prompt by producing both sides of the conversation. Note however at some level the training has encoded enough structure through compression that it is able to simulate all sorts of phenomena, from human conversations to code. The great R&D difficulty here is to scale pretraining so that it can proceed smoothly in vast distributed datacenters in a fault-tolerant manner.
The next few stages are collectively called post-training, and typically consist of supervised fine-tuning, then reinforcement learning.
In supervised fine-tuning, the model is further trained to predict the next token, but on a much more focused data set of natural language conversations where the "assistant" and "user" turns are explicitly delineated with special tokens. The output of this stage is a model which is capable of carrying on proper conversations, but typically with no ability to creatively problem-solve, and less of a personality. The data and compute are many orders of magnitude smaller than in pretraining.
The reinforcement learning stage used to be a small part of model training, but ever since AI-assisted coding took off, it has become larger and larger chunk of training. In recent models, the compute spend on RL has allegedly come to rival or even exceed that of pretraining [1], which is a bit scary because RL is classically what lead to sci-fi like AIs which are extremely good at accomplishing goals to the detriment of everything else.
The way that RL works is that you put an instance of your model in some environment (such as a VM containing a git repository) and give it a task (such as fix the linked github issue). The model will then generate a bunch of attempts to solve the task which we call "trajectories", in most cases there is either an objective measure of the task success (such as passing the tests), or a fuzzy measure (such as having another LLM look at the results and provide a score). This is called the reward, and the model will learn slowly by producing trajectories that receive reward. It can actually be quite hard to prevent "reward hacking" from the model here and the rewards must be shaped very carefully, much R&D labor goes into here, as well as similar challenges to distributed pretraining.
A significant challenge is that coding/knowledge work tasks these days are getting extremely difficult, we are far beyond 2024 days where models could barely solve the easiest problems in SWE-bench. Tasks at the frontier now look more like mini projects that would take humans multiple hours or even days to finish (or in some cases, research-style tasks that would be beyond reach for even top human experts, such as the Erdős unit distance problem which was posed in 1946 but wasn't solved until recently, by GPT-5.5). Huge amounts of trajectories must be produced, and huge amounts of them produce zero reward and therefore are useless for learning. Getting a cold start requires running tens of thousands of instances of your model in VMs in parallel for multiple days to produce trajectories, to say nothing of the GPU costs.
So what do you do when you only have a model which is capable of basic conversations but cannot even begin to tackle basic coding tasks, use tools, etc? The approach that companies behind the frontier have decided on is to bootstrap their learning process by having an already extremely intelligent model such as Claude produce hundreds of thousands of seed trajectories for them. Then they can use this data to get a warm start and begin learning immediately. And if you use Claude for your reward model too, you get to skip the nastiness of reward shaping.
Therefore, even if in number of raw tokens the data are much smaller than internet-scale pretraining data, the value that each token provides is far far greater.
[1] For example, Grok 4 compute spend on RL was ~100% of that of pretraining: https://www.interconnects.ai/p/grok-4-an-o3-look-alike-in-se...
props for a great write-up
Actually it's a hit piece.
A description that highlights the importance of RL is a hit-piece?
Training isn’t a single homogeneous step. It starts with pretraining which requires bulk PB of data but you have less quality concerns here. You cover the whole data distribution. Later stages require less and less but increasingly higher quality and complex datasets. The late stage ones are highly curated and might even be sourced from world subject experts. This is where frontier labs with big pockets have the advantage.
Actually nowadays LLMs are only trained with TBs rather than PBs of data, and it's not too hard to find GBs of agent traces online.
This might be like an observational study vs a study with a control?
From what I understand, at this point, the main value of stronger model outputs is simply to bootstrap reasoning behavior during the RL post-training phase. It gets you past the “cold start” problem with RL, after which the outputs aren’t needed anymore. From then on, it’s hill climbing and that requires environments for the model to interact with get rewards from.
> But if you show them a jailbreak of their model that bypasses their safety, they'll tell you that any model can eventually be jailbroken so don't worry about safety.
They claim two things:
1) The specific, available jailbreak for Fable 5 is not dangerous - this has been confirmed by multiple experts, and there is no credible evidence against this claim (in other words, Anthropic is probably correct)
2) It is impossible to build an LLM that is immune to all jailbreaks. Again, there is no credible evidence against this claim, i.e. Anthropic is again entirely correct.
If #1 was false, they could just publish the details of the jailbreak - it supposedly only works on Fable 5, so there's no possible danger.
If #2 was false, surely some other LLM lab would have done it by now. Especially since a number of governments have made it clear there is a market for such a project.
Supposedly the details of the ‘jailbreak’ are that you give it insecure code and say “fix this code”, and it does, and then you ask it for test scripts and that’s effectively an exploit against the unfixed code.
If true then I have no idea how anyone’s going to release a useful model that doesn’t have the same jailbreak. https://www.theregister.com/security/2026/06/15/feds-freaked...
If that's the extent of the jailbreak, then the government should have banned every existing LLM - their story only makes sense if there's some Fable-specific capability that got unlocked.
There’s no logic to it, blocking fable was retaliation and market manipulation by the current admin, nothing more. Poorly conceived as well.
> If #2 was false, surely some other LLM lab would have done it by now.
This is a logical flaw. LLM that is immune to jailbreak _could_ exist, but not yet, or maybe nobody talks about it. Yes there's a market, but all of these AI boom is too recent to make any claims.
Like how would you even define what a jailbreak is?
I think pretty much parallel to how social engineering, manipulation, scams work. LLMs are being trained to have human values, prioritizing human lifes, yet people are shocked it will spurt out how to make a nuclear bomb because grandma is being tied to a train track as a hostage.
I would also spurt out how to make a nuclear bomb (ie public information you can find using google) if I was told that's what I gotta do to save grandma tied to a train track as a hostage. "AI safety" is such a shit show.
I'm pretty sure that Gödel incompleteness theorem and its consequences pretty much guarantee #2
I'm guessing you mean, the incompleteness theorem guarantees that nobody can prove their model is un-break-able?
I don't think that's quite what it means. The theorem says that it's impossible to write a function, "will_halt(program, input)", that will be correct for all possible {program, input} pairs. But for a particular program, you may be able to write a proof that it will halt for all inputs -- that's what software verification is about.
The implications here would be that nobody can create a "will_jailbreak(model, input)" function which works for all model/input pairs. But we don't need a general function which works for all model/input pairs; we just need a way to prove that for a specific model, there will be no jailbreaks for any input. As with software verification, this may require that the model be developed in a specific way.
Granted we don't currently know how to make such a proof regarding neural networks; but that's not because of Gödel.
Mind to elaborate?
No actually I don't think it does and I don't think they're related.
Exactly. It's impossible to guarantee #2 doesn't happen (ie protect against all jailbreaks) for any system of sufficient complexity.
If you’re doing evals, you’re basically doing RLAIF without training a model; just looking at the results.
Fundamentally it is very difficult to stop this while still making your AI models useful.
Similarly, if you did a corpus study on bioRvix to summarize recent science findings — you could use the same questions and answers to fine tune a model.
There is no way to communicate information at scale to companies through the API, for anything approaching a real application, without that information forming a corpus another model can be trained on.
But it wouldn’t be the first time they broke a model:
Their “guardrails” that cause it to reject user prompts also means it relies on its pop science summary of medicine to tell you why bioRxiv is wrong rather than accurately summarize the papers.
They’ve successfully created a smug, argumentative average of the internet which refuses to even consider it might be wrong or that it’s reading a science paper which is based on measurements and not vibes — but why would I pay for that?
I get it for free online.
Doesn't "real" distillation use the logits instead of the final tokens? I would classify this more like using a model to generate synthetic training data.
The compute deficit of Chinese Ai companies is real, and it IS THE ONLY competitive advantage that Western companies have.
The only way the U.S. keeps that edge is to prevent distillation. The only way Chinese companies can make up for the deficit in compute is to distill. There innovation in great supply on every side of the Ocean. Its about the chips. And in terms of national security, for the U.S., and for China, its about the chips and the distillation that undermines that advantage. This is an arms race.
If compute or access to training data were the only issues, then companies like Meta and X.ai (Grok) should be doing better, even Google for that matter. Musk even admitted that Grok used training data from OpenAI models.
https://techcrunch.com/2026/04/30/elon-musk-testifies-that-x...
While there is no moat as such, there is still a lot of expertise that goes into training SOTA models. There's a reason Google was willing to pay $2.7B just to get Noam Shazeer back to improve Gemini.
You got that wrong. The forcing function of compute scarcity is an advantage not a detriment. The amount of investment pulverized in performative model training and dead ends (Hi Sora) should make this obvious.
If saying “plz don’t distill me” is your moat, you don’t have a moat.
No. What will happen is it will turn dark. No public release. National Security uses only, or in carefully vetted industry settings.
There’s huge issue with that approach. It’s not a multi trillion dollar business.
yeah nope, it won't happen, the snowball is already rolling
Good luck not crashing the markets and the economy.
And good luck not staying behind when you can't monetize your gargantuan investments and have little incentives to make your models better as the world moves on.
you mention wedding ring like it's a bad thing
Define compute deficit?
They've been bringing out open weight models competitive with frontier models. How could they do that if they had a compute deficit?
If they need to divert inference resources to train models, this counts as a compute deficit to me.
I'm using GLM-5.2 daily for my own stuff, and during Chinese business hours, specially on their afternoon, it's a festival of rate limits.
I believe this article is about the technique they may or may not have used.
> The only way the U.S. keeps that edge is to prevent distillation.
For how long ? year ? how long till model that is year behind will be fine for 90%+ use cases ?
Putting aside agentic coding, that is to say, if you judge LLMs as a consumer technology (an old-fashioned idea for the inward-looking tech industry admittedly), then open weights LLMs, even quite small ones like Gemma 4, can likely already satisfy 90% of applications with a bit of help from search and browse tools.
Much of the arms race for better LLMs exists to satisfy only the IT industry's needs.
Yeah I think the technical term is something more like “pseudo-labeling”. The OG distillation requires logits which Anthropic doesn’t provide.
I've used RLAIF to build out heuristic based non-LLM models for various decision systems and achieved like, 95% F1 on certain projects. We're in a place where models can be used to fine tune a lot of stuff via loops.
> These complaints of distillation are inflating the problem to make it sound worse than it is
This is, in part, a problem every judicial and legislative system has faced since forever: form versus function.
Take a classic elicitation spying techniques: a foreign spy meets a military officer/scientist at a bar, strikes up a conversation, makes an observation wondering how could a missile hit some target at some accuracy and elicits a response that with laser guidance it is entirely possible. From this they get info that there is some technology to laser guide missiles. Or in retail, a competitor hiring a secret buyer for core baskets of goods and analyzing prices in the receipts.
The function is espionage, the form is conversation and all info is in a sense provided willingly. Where do you pull the slider?
These distillation "attacks" are not only indistinguishable from evals, they ARE evals. The function is own model training, the form is eval. Normally, one would expect to have risk benefit analysis based discussion which direction to push the legality slider to. The problem with these recurring statements is that they invoke enshitification of legislature.
Chinese labs access Claude via API. Isn't it the black box method by definition?
I'm sorry, but you got the terminology exactly backwards. Training on the answer is called supervised fine-tuning.
Just for the sake of clarity:
0. Full distillation uses logits of the teacher model - that's much more information than the text itself. This is a kind of distillation used inside labs, but one can't distill Claude this way as logits are not available via API.
1. Supervised fine-tuning on synthetic data might be called blackbox distillation. I guess that's what you meant in your case (1).
2. Reinforcement learning (like RLAIF) uses least amount of information from the teacher, i.e. only few bits per task.
>But if you show them a jailbreak of their model that bypasses their safety, they'll tell you that any model can eventually be jailbroken so don't worry about safety.
Yes this is in line with what Anthropic said in their public statements about their Fable access restriction by the government directive. The hypocrisy and inconsistency in their statements and behavior feels quite childish and controlling. I believe our companies and their leaders, friends among our other influential leaders and leaders from rich social classes, want to actively hurt most people as this behavior looks to be quite self-interested.
Not to mention, the person who brought this quote unquote jailbreak to the Trump Administration was Amazon’s new CEO. They know their IPOs are coming up, so locking their competitors out of the U.S. (even if just for the weeks surrounding the IPO date) would be a major boon. The White House seems to love making announcements just for the sake of making the market move…. Coincidentally, right after POTUS buys a massive amount of the benefactory company’s stock (Buy Dell Computers, lol)
Can you reach into the model and "transplant" weights directly?
I'm not 100% sure it's not possible. If (I don't know) it's possible to freeze the temperature of the model so it's deterministic, and if you could make a map of produced words back to tokens (via HMM probably), then you can probably alter a minimal input and observe the output to model it. If you perform waves of such minimal alterations, you can expect to be able to locate the distance where each alteration impact the model (the idea being that a small alteration on output is likely due to the last layers of the models, and a small alteration is likely due to the deeper layer). Once you've located most of the last layer(s?) weights, you can try to solve for them. With a hundreds of billions weights model, the last layers will likely be so huge that it's probably unfeasible technically, but it's theoretically possible.
No, you'd need to have the model on your filesystem for direct access, and then the architecture would need to be the same.
If you have access to the weights, you can just use them as is...
Anthropic are not saying they have been hacked - they are saying that Alibaba have been sending lot of requests to their servers.
You can do things like that - one example is averaging weights between related models - but not with Anthropic's models, because outsiders don't have access to the weights.
Weights are just data a server, so we don't know outsiders have access (either via breakin or arrangement).
Yes, obviously. That's not the point.
> These complaints of distillation are inflating the problem
They’re also missing the point. What would have happened to a member of the Manhattan Project who, through personal pursuit of profit, neglected their duty enough to let the bomb leak?
The companies are all for-profit companies, its not like they're selling out some national security goal for profit, profit is the point.
Anthropic already heavily restricts Chinese traffic but that only jams up researchers and regular Joes. Anyone motivated enough can hop a flight to Singapore with an nvme drive in their pocket.
Chinese companies are engaging in anti-competitive practices, as usual. They are rogue actors on the economic scene. If it were feasible, they'd be widely banned, and for good reason.
Bringing more competition is "anti-competitive" now.
Merely copying products that actual companies produce and making them cheaper is anti-competitive. There's no incentive for the products to be developed in the first place in a market if this is happening. This is why copy protections exist in civilized countries (not China and to a lesser extent India).
That's why IP was invented, but what IP are they infringing? Not patents, not copyright, not trademarks, so what? Making something cheaper than someone else isn't anti-competitive. There are a lot of businesses that do that. That's the very essence of competition.
Here's what is happening:
Chinese resellers are offering Claude tokens at 70-90% below official Anthropic API prices. They achieve this by reselling capacity from pooled Claude Max accounts, payments fraud, and also reselling the model output & reasoning chains to various Chinese labs. They are subsidizing model access in exchange for user logs and reasoning traces, which they then sell as training data, allowing them to operate below cost.
Claude and ChatGPT are both blocked in China. You need to use a VPN to access either, and you can't pay with a Chinese bank card. So most people who want access to Claude buy access via a reseller. It's the easiest and cheapest way to access Anthropic models in China.
These resellers operate tens of thousands of bot accounts, which is also why Anthropic introduced identity verification, to slow down the onslaught of bots.
Here's one token reseller, they're offering Opus 4.8 at a 93% discount below official API rates: https://yunwu.ai/pricing?provider=Anthropic
This is one reason why DeepSeek & GLM are priced so cheaply, they are competing with impossibly low token prices in China. They have to keep prices low, in order for people to use them.
I shared this story a few months back, but it never got any traction. It explains the token resale economy in China, it's an excellent read https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens...
> This is one reason why Deepseek & GLM are priced so cheaply, they are competing with impossibly low token prices in China. They have to keep prices low, in order for people to use them.
This one does not make sense to me at all.
Deepseek and GLM are openweights, even US inference provider are selling them at much cheaper price. The price is cheap because the model is more efficient.
DeepSeek permanently cut its V4-pro API prices by 75% because they were too expensive. Without the price cut, Deepseek V4-pro tokens would have cost more than resold Opus 4.8 tokens.
Opus 4.8 is a more capable model, so almost nobody was going to pay for V4-pro at the original price.
> Without the price cut, Deepseek V4-pro tokens would have cost more than resold Opus 4.8 tokens.
You mean it's functionally as if American tokens are being price dumped in China and Chinese model providers are being forced to compete with that and innovate? So many delicious layers of irony, lol :-P
China also have trust issue with American companies. Most of State-owned companies will not use those services even if they can directly access them.
And? The US feds wont allow even local Qwen or Deepseek models either. "Evul godless commies" or some such nonsense.
If other providers can match Deepseek's first party prices, that probably means that the economics for running inferencing work out for them.
Urm, no? I man they did cut prices by 75% that part is true - but they reduced a starting price that was below sonnet.
Also it's a open weight model, doing that is impossible long term because the real price will be set by the other model providers, who priced it around 60% of sonnet inference cost. Had to look that up though, so that's today's pricing.
Is there a contradiction here? If resold Opus tokens are sold at a 93% discount, you can be a lot cheaper than Sonnet while also a lot more expensive than resold Opus tokens.
I see, After rereading the comment I was responding to I realized I probably misread/misinterpreted what they wanted to convey.
I think there isn't a contradiction and I was just confused. The price may have been discounted only to get below the price point of opus resellers. I do not have enough information on that to make any clear determination on that topic.
It's somewhat difficult to have any sympathy for Anthropic here. They're entirely responsible for selling tokens at below cost, with the age-old bait-and-switch tactic.
If they weren't doing so, then these Chinese resellers wouldn't be viable. Radical idea, but how about they actually charge a viable price, even on subscription plans?
If resold Anthropic tokens undercut even the at-cost open-weight model tokens, because they're reselling subsidized subscription tokens, then you'd have to start selling open-weight model tokens at a loss in order to match them.
Also, wouldn't that claim only hold in China?
I'm an European and I'm not using those proxys the article describes.
>They achieve this by reselling capacity from pooled Claude Max 5x accounts, payments fraud, and also reselling the model output to various Chinese labs.
>Here's one token reseller, they're offering Opus 4.8 for a 93% discount below official API rates: https://yunwu.ai/pricing?keyword=claude
But is it cheaper than getting your own account? Otherwise this sounds like the "anthropic/openai are losing gazillions of dollars because they're selling $1k worth of tokens for $100" line that's commonly trotted out by AI bears.
It's very difficult for people to create personal Anthropic accounts from China. Anthropic blocks Chinese bank cards, so people must pay with a foreign bank card, which they likely don't have. And even if they manage to set one up, they have to access it via VPN, which eventually gets the account flagged. They then have to complete identity verification, which most Chinese users are unable to pass.
There's a similar Claude resale market going on in Russia. On Funpay they are selling Claude tokens for roughly 20-30x cheaper than official Anthropic API pricing.
And phone number verification too? So that's 3 hurdles to jump to just get opus.
for verification you can buy phone number for $1 easily.
Doesn't ~every phone number verification service check the telephone provider and only allow from a select whitelist of residential providers?
You can easily get lots of esims for $5-10 months in EU / UK or just for free.
With esim and virtual credit card services, why would that be a blocker? Who's got a landline these days?
> Claude and ChatGPT are both blocked in China
So it's presumably cheaper than attempting to spin up your own method of circumventing the blocks.
You can use it as an API unlike the subscription.
Maybe these resellers are using stolen American credit card numbers? Reselling Claude access seems to be a nice way to launder the money.
This is great for competition! Chinese vendors offering a cheaper solution = what economics told me the free market was all about.
I also learnt that Anthropic should get better at what they do if they want to compete. If not, somebody else will win.
Or does this not apply to huge US corporations any more?
China aren't offering a cheaper solution. They are subsidizing an existing one (which is already subsidized) in order to gain foothold. The difference is that in the US subsidies come from VC, while OP implies subsidies come from the AI labs that buy the training data (which may as well also be VC backed, so just one extra hop).
This isn't "the market working as intended", this is an exhaustion fight to the bottom where the one with most money gets to stay in the market. As with most venture capital startups. I believe this VC tactic is a well documented "cheat code" to bypass market forces and build a monopoly. I find it hard to compare that with a free market.
However, I don't really mind China "stealing" from Anthropic. For us consumers we are getting the cake and eating it too. I.e we are getting rapid improvement to the tune of over a hundred billion dollars in funding, yet the market remains big enough that there's a chance of it not ending up as a monopoly in 20 years. And venture capital are footing the bill. A part of their investment is practically being redirected to fund Chinese AI development. It lets us live out our lives as happy CAC farmers[1].
So I would argue its not as much of a "cheaper solution" as it is intentionally and maliciously abusing another company's product to extract more value than the billing plans intend (given an average user), and further subsidizing the product by selling this data to competitors. But I don't necessarily think its a bad thing for us end-users. Nor for the market. But it is bad for Anthropic and its investors.
[1]: https://phrack.org/issues/71/17
> China aren't offering a cheaper solution. They are subsidizing an existing one
Chinese labs are also pursuing legit frontier-advancing R&D into efficiency and publishing papers in the open, a culture that's in retreat at top American AI labs
Their is plenty of innovation happening on both sides of the Pacific. Again, China publishes open source because they don't have another game they can play. They distill because they don't have the compute to compete. They are great lab, for sure, but the fundamentals are driving their behavior.
The fact that are people that genuinely believe you can train an LLM by using random QAs obtained from another LLM is astonishing. Let alone the fact that it makes absolutely zero financial sense.
At this point this is being repeated so often that completely uninformed users are taking this at face value.
Oh yeah. Strategic disruption technique or not its a breath of fresh air.
> China aren't offering a cheaper solution. They are subsidizing an existing one (which is already subsidized) in order to gain foothold.
In my economics classes, we were told that (in a "free market" argument) the best thing to do if a subsidy is making something you want cheaper is to use it. You're getting your thing, and at a reduced cost.
(I'm not really replying to you per se, I'm curious how "free market" folks in these comments would respond to this.)
This is why I don't understand why people complain about impractically cheap Chinese solar panels. The rest of the world should buy enormous quantities and bankrupt the mofos and hugely benefit along the way. Then later they can set up their own solar panel industries.
Because they arent selling at a loss. The business pipeline is subsidized by the state. But end to end from mining the minerals to shipping you the solar panel everything is "in house". Its all in China. Thats why why can sell so cheap. Its even cheaper to make.
This narative that the CCP is just subsidizing all business to "beat america" is just dumb. Its the build process being made cheaper by the government. Not the final product.
Let’s not act like the US government isn’t subsidizing AI either with massive contracts. Anthropic is selling subscriptions at a loss; reselling tokens is just arbitrage.
The US government subsidizes US AI labs via incentives.
The CCP is the Chinese AI labs.
I am not aware of any US government AI labs (besides perhaps a small spattering of national lab research or the like)
There is very large difference that your either need to be poorly informed or purposely driving an agenda to miss.
I’m sad for you that propaganda has destroyed your critical thinking ability. Qwen is an Alibaba product, Deepseek is from a private Chinese hedge fund. Those are not the CCP. The Chinese economy has vastly evolved in the last decades. The CCP doesn’t have any more special control over the Chinese labs than the US has over our labs. The White House can do whatever it wants to keep labs in line. Fable just got pulled because the US gov ordered Anthropic not to give access to any foreign nationals, including Anthropic employees.
Please, show me one Chinese court ruling against the CCP and I will believe you. Anthropic can go to court and have the order overturned if it isn't legal (with legality being born from elected representatives), it happens all the time.
Just because Xi Jinping lets companies play mock "private businesses", does not mean there actually is private business. At the end of the day, the CCP still has final say in everything, and Xi has final say in the party. There is no constitution (in the US judges swear to the constitution, in China they swear to the party), and there is no balance of powers.
It's just one guy, running experiments the way he see's fit.
You mentioned propaganda, take heed.
How many court orders has the US govt completely ignored? Do the bought and paid for members of SCOTUS ever rule against Trump or their donors?
> It's just one guy, running experiments the way he see's fit.
This is moronic. Exercising a degree of control does not equate to making every decision and running every organization.
Well they shot down his landmark tariffs for one, and the next court ordered refunds so...I'd be delighted if you could share cases of the government ignoring court orders (not to be confused with challenging them, like any functioning legal system has).
Also, obviously Xi doesn't make every decision. No dictator ever did that because it's impossible to do. The distinction is that no one has ever (or has the ability) to over rule what Xi decides. So if Xi has a stroke and wants DeepSeek to start manufacturing underwear, they will be ordering sewing machines tomorrow. Any sense of "private" is a farce.
As far as I can tell, the Chinese government itself is complaining about 'oversupply' in the solar panel market. Ie it doesn't sound like they are subsidising it anymore.
The Chinese government has stopped direct subsidizing solar panels years ago. I think it was around 2019? This resulted in a lot of companies going under at the time.
It did not stop solar panels getting cheaper and cheaper because of the whole integration and mass production (with healthy free market competition).
The last subsidies like export value-added tax rebates for solar panels and lower rebates for batteries are ending in 2027.
China their main power is, the ability to have everything inhouse. Yea, they subsidize a lot of stuff until it hits critical mass, and then you have often a healthy industry with lots of competition.
China alone has like a few 100 car manufactures because of the subsidies, and over time there will be consolidation / buyouts etc but the end result is a healthy new industry that exports. With again, everything internally being produced.
This is why our subsidies fail. We do one sector, often a few companies at best. This results in few competitors, expensive prices, and often reliance on externals that can bankrupt those companies. And que how we wasted again dozens of billions in propping up a industry with no competitive edge.
People can cry about China but they are actually doing work, despite the mass amount of corruption. That is the big difference with here... Mass corruption got in the way of national security, plop, people go to jail. Industry quickly gets their ** together. Here ... give billions, and the money vanishes, with no real consequences.
Selling at a loss isn't required.
Local governments are over-funding numerous producers (though cheap loans and other subsidies and incentives) creating excess competition. This is an ongoing problem and is a huge misallocation of capital. Increasing demand just drives this process harder and puts downward pressure on margins. As soon as they try raising prices, or just through satisfying total demand, demand collapses and they (almost) all go out of business.
The Chinese model has weaknesses, we should be exploiting them.
the fact that you think an organization that pulled 300 million people out of poverty in 20 years with strategic planning and a controlled economy has this not covered is mind blowing. they killed the made in USA slogan in less than 40 years. they'll be fine.
The textbook poverty which are created by such organization itself with strategic planning and a controlled economy in the first place, killing ~30 million. All the more impressive it only took them under 10 years.
Most everything they teach in economics class is wrong. You would be better off ignoring everything they teach you.
Could you elaborate?
> China aren't offering a cheaper solution. They are subsidizing an existing one
So basically like US companies subsidizing offerings with selling user data, ads for crypto scams, manipulation for elections, making people addicted to gambling and so on?
Seems fair and an improvement as you can choose between that and not. Unlike say offerings from Meta where the data selling and efforts to further gambling addiction is always included.
Which part are we supposed to have an issue with? The selling data to offer cheaper compute? Products taking over markets with below cost pricing because they have money and ruining the free market?
Because all of that is considered totally okay when every single US big tech company does it.
All I can say is lol. DeepSeek showing 3 order of magnitude efficiency gains over the performative capital furnace that was training and inference absolutely moved the bar here.
Chinese models are years ahead of american models on multimodal comprehension, better yet,they publish on what makes the models tick and release weights openly.
Chinese research outout, publically released, has also contributed in big ways to features present in every single US model. Yours is a bit of an unfair take I'd say.
Besides, claude will think its chatgpt sometimes, so clearly this isn't a problem restricted to china, turns out unethical companies will do unethical things /shrug
A trillion dollar ipo jist occurred for a company whose main line of business is almost entirely subsidized by government contracts
Is buying launch services really a subsidy, or did I misunderstand?
> This isn't "the market working as intended", this is an exhaustion fight to the bottom where the one with most money gets to stay in the market. As with most venture capital startups. I believe this VC tactic is a well documented "cheat code" to bypass market forces and build a monopoly. I find it hard to compare that with a free market.
Why? Lots of people try this tactic, but hardly anyone ever succeeds. Meanwhile, the customer benefits.
Lots of people have succeeded. Neither Anthropic nor OpenAI has any technical advantage in the field of subscription engineering.
Please give me a few examples of people succeeding with the technique.
Specifically, examples of people later exploiting their monopoly to charge people more than they otherwise would have paid.
Amazon and diapers.
Ubers have become pretty expensive these days.
Well, is Uber a monopoly and collecting monopoly rents?
I use Lyft because my phone is too old to run the Uber app. This says nothing about market share obviously.
> This isn't "the market working as intended", this is an exhaustion fight to the bottom where the one with most money gets to stay in the market.
That's, uh, pretty much exactly how oligopolistic markets function.
> I find it hard to compare that with a free market.
Well, to have free market you need to remove as much barriers to enter the market as possible. Huge capital investments required for entry and intellectual property laws are two examples of such barriers. Subsidies kinda supposed to help alleviate the first one.
I mean, for what it's worth, we have subsidized Anthropic by allowing them to train on copyrighted stuff. (I know it is still legal, and I support the legality, but the economics are what they are with people's content paying a big one time subsidized cost (to the level of at least 500B).
So, the least Anthropic can do is pay it forward.
I am mostly economics illiterate but I understand a subsidy to be an economic concession given by the state to an entity which gives said entity a relative advantage compared to its peers.
In that sense (which could very well be bogus), letting a company violate individual IP of basically every human is less of an economic concession and more of unconsented to IP open season.
Even if one were to drop "economic" from "economic concession" and instead view a subsidy through the lens of a more general concession, one could say that the US Govt gave US AI companies a legal concession to sidestep the copyright protections of other US entities. But the US Govt should only get to undermine the copyright protection of other US entities - who gave American companies the right to violate the copyright of non-Americans?
That's some "download a car", $100000 per infringement pricing logic. No one is paying anyone 500 billion dollars. I'm sure rights owners wanted that, and more too, but it's nonsense to call it a subsidy that they didn't get it.
If we as individuals were sued it surely would be at least an order of magnitude difference between what is required from us vs Anthropic or OpenAI. That’s even completely ignoring the marginal utility of money. It is absolutely a subsidy. It’s just less fair because that power, to pay pennies on the dollar, is only given to corporations.
doesnt VC money subsidise stuff all the time? Isnt that how Uber and AirBNB undercut competition?
The VCs footing the bill is really your pension funds and 401Ks and banks passing through the VCs. If VCs lose money the contagion spreads through the economy.
>This is great for competition! Chinese vendors offering a cheaper solution = what economics told me the free market was all about.
Yeah, like all those Chinese bootleggers selling DVDs for a few dollars rather than $20. Free market!
https://news.ycombinator.com/item?id=48664814
"Information wants to be free"
Anthropic profited from training its models on all kinds of copyrighted information, live by the sword, die by the sword...
Their model weights, training data, training methods, etc are all going to leak to China over time.
Nobody on a site named _Hacker_ news should be all that upset about this.
Seriously AI companies complaining about fair use is the biggest case of crocodile tears I can think of. Irony has been dead for a while, but they dug up the corpse and set it on fire anyway.
Totally agree.
Don't forget insider threat vector, too.
I deliberately didn't mention any threat vector.
I would assume China is working on liberating Anthropic weights through the battle-tested strategy of finding someone in a privileged position and getting them laid, etc.
You’re mistaking the original term hacker, a tinkerer of systems, for the black hat variety.
Hackers didn't use to spend a lot of time defending trillion-dollar corporations and their intellectual property rights.
black and white hat is relative. someone breaking into a state run database in a dictatorship and stealing documents that prove some opposition leader was murdered would be a black hat criminal if you ask their government. a hacker jailbreaking a phone to let people fix it without expensive official service is a black hat to the company. we should really switch to saying offensive and defensive or something else that doesnt come with moral implications. maybe lawful and chaotic.
I suppose his point was that the both parties are black hats.
What true hackers really did was discuss the definition of the word and how to use it
There is no real difference
It’s true that the meanings of words can change over time. Whether or not that’s a good thing is another question entirely.
Yes there is.
Care to elaborate on your side or should we just leave it there?
there is, the original hackers built thinks, they didn't attempt to destroy or coopt them
https://en.wikipedia.org/wiki/Blue_box
Bootlegging is copyright theft.
Is Claude output copyrighted?
If anything, a tremendous amount of Claude’s input is copyrighted.
If there’s any bootlegging going on it’s Anthropic that’s doing the bootlegging but having mirrored the video etc sufficiently to beat copyright law.
>Bootlegging is copyright theft.
Ok, but what about those shady sites that resell Windows education keys? They're certainly a "better experience" than buying legit keys, by virtue of being significantly cheaper. You aren't even really committing copyright infringement in the process, because Microsoft gives out windows isos for free, and the seller is really selling a random 25 character string, which can hardly be copyrighted.
>If there’s any bootlegging going on it’s Anthropic that’s doing the bootlegging but having mirrored the video etc sufficiently to beat copyright law.
US courts have consistently ruled it's fair use.
>>If there’s any bootlegging going on it’s Anthropic that’s doing the bootlegging but having mirrored the video etc sufficiently to beat copyright law.
>US courts have consistently ruled it's fair use.
And they also have ruled that the that output of an AI isn't copyrightable.
As such copying claudes output isnt even fair use as that is an exemption to copyright but the same as copying public domain work which any and all are allowed to do.
> because Microsoft gives out windows isos for free,
… with a license that only allows you to use it for certain purposes, subject to certain restrictions.
> and the seller is really selling a random 25 character string, which can hardly be copyrighted.
1. Copyright is about creative works. It is possible to have a meaningful creative work no more than 25 characters long (or equivalent). Music is particularly good at this.
2. The key itself is not copyrighted (it’s not a creative work), but is reasonably interpreted as a copyright circumvention device. See also https://en.wikipedia.org/wiki/Illegal_number.
> US courts have consistently ruled its fair use.
Like Adam Smith wrote in The Wealth of Nations “‘Free market’ is when a company receives a favorable ruling about copyright in the United States”
> Ok, but what about those shady sites that resell Windows education keys?
Yes, they are fine? They might no longer include full first party support by Microsoft for not being "new". Same as buying a used car (also comes with the "shady sites" for a far longer time).
Though this not making any difference by Microsoft not doing any support either way to make more money is a business decision by Microsoft.
What about those store brand cereals? “Chocolate puffed balls” for a fraction of the price of Cocoa Puffs™?! You all may laugh until your waterways are under siege and you find Cap’n Crunch™ keelhauled by thrifty shoppers
and Chinese courts are ruling that using Claude is fair use.
american courts have ruled that theres no copyright at all on LLM outputs
The current case law in the US is that the raw output of an LLM cannot be copyrighted without further meaningful arrangement or alteration by a human author.
I think renting out ID to let others in without telling the admin is generally unlawful in many places
The output of Claude is not eligible for copyright protection. I'm not sure how the analogy of bootlegging DVDs would work, given that.
I suppose you are violating the TOS by reselling a service, even if the output can't be claimed as belonging to anyone.
Sure, and Anthropic is allowed to cancel the licenses of people doing that, as they do
You can write anything in TOS, many parts of it is non-enforceable and depends a lot on the local laws.
Free market would of course allow bootleg DVD sales, state regulation that gives monopoly rights restrict it.
In the context of LLMs, monopoly rights haven't been created (yet anyway).
Fun fact: for a period the US (or american colonies) didn't have copyright but Europe did, so people could copy and sell English (and other) books for free.
BigAI are all in the bootlegging market themselves, so it's always funny to see them complaining about others copying their "product".
It's quite curious how multi billion dollar enterprises can't compete with a Chinese bootlegger with a big jacket, tbh.
Imagine having such a warchest and being so bad at business, lol.
Bad at business? One of them has to make the thing.
Maybe the making of the thing should be paid for before it's made, rather than hoping that selling copies will recoup the investment. I.e., go back to patronage while abolishing copyright.
This doesn’t support your previous theory.
This type of "resource curse" paints a perfect picture of why US based frontier providers are set up to fail. They want, and have, few restrictions and along with that unlimited warchest. The Chinese on the other hand aren't burning billions like millions. Anthropic, OAI, Google, Meta... They're all phenomenal examples of waste, corporate greed, inefficiency and are the reasons people hate tech bros at this point. Whining and crying from their super yacht, Parkinson's law is alive and well!
My biggest concern with pirating has always been malicious programs. But companies still need to show value in their products or people will pirate.
What added value can Anthropic give users not available to pirating users? That is what they should ask themselves.
ZDR but that is meaningless if the person wants to do nothing more than cheat on homework (or has enough hardware to run a local model)
any third party provider can offer zdr. if its a reputable company in a place like switzerland or germany i would trust them more than anthropic to hold up that promise.
> Yeah, like all those Chinese bootleggers selling DVDs for a few dollars rather than $20. Free market!
It's supremely ironic analogize distillation to copyright infringement when it's literally what Anthropic was found guilty of. It's not illegal to distill. It is illegal to pirate. And it's what Anthropic was found guilty of, not Alibaba.
https://apnews.com/article/anthropic-authors-copyright-judge...
I bet you've downloaded a car.
And those darned printing presses distributing works that were written prior to their existence.
This is also a good thing fwiw.
More like one bootlegger complaining that another bootlegger is copying their bootleg DVDs.
Except Anthropic didn't produce the movie.
So it's more like one bootlegger sold the DVD for $20 and their competitors are undercutting them for $1. Who's the bigger thief here now?
Capitalism as intended!
I get the vague impression that this was written in a sarcastic way, but it has a straightforwardly true literal read because yes, this is what the free market is about and Anthropic will have to compete with the Chinese if they want a big share of the market. Chinese models are cheap and good; even without reselling Anthropic's services they're competitive. Which reading did you intend?
And, gotta say, the idea that the Chinese are better at selling US models than the Americans is hilarious. There might be an economic study here somewhere about just how anti-consumer and anti-progress their IP laws turned out to be. We've got an entire postindustrial revolution centred around who can ignore the most stupid laws.
> the idea that the Chinese are better at selling US models than the Americans is hilarious
This is not the right deduction.
China blocks foreign AI from operating there.
> China blocks foreign AI from operating there.
Given the current US government's tightening of export control restrictions and the introduction of a bipartisan bill to block use of Chinese AI in federal agencies, I'd say the two countries' positions are not far apart.
https://apnews.com/article/ai-china-united-states-competitio...
Yes neither are free markets
I think you will find that it's the USA government imposing such restrictions.
That is ALSO happening, but that's beside the point.
Chinese AI apps like DeepSeek are freely available for ordinary Americans to download and use. There's no federal law banning private citizens from using them.
So to claim that Chinese companies are better at selling American companies' work than the American companies can do themselves when they are prohibited from operating in that market, is the wrong deduction to make.
there will be soon enough. TikTok is the example for the US clamping down on companies that dont toe the regime line on israel
> Or does this not apply to huge US corporations any more?
When it comes to favorite companies of the tech communities, it's almost always "Rules for thee, but not for me"
The standard stance is "they can do no wrong and they are absolutely perfect". I mean, look at any thread with anything about Apple in it.
> what economics told me the free market was all about.
Don't complain when US starts to play by the same rules China has been using for decades.
What is the implication here? Are you warning that US corporations might start doing something shady, like scraping the internet at large scale for training data? Or mass-dowloading pirated copies of books, completely ignoring copyright?
I find it hard to imagine a future where US corporations have degraded to such a point.
Or building backdoor in to the physical servers sold around the world?
No, he means that the US will close most of its domestic market to competition just like China has for decades, and the US may start subsidizing and dumping its goods everywhere
> the US may start subsidizing and dumping its goods everywhere
Isn't that exactly what companies like Uber have already been doing? Take VC money, sell goods & services at a huge loss, wait until the competition goes bankrupt.
Exactly, it's funny how most Americans have no self-awareness on this topic.
And beyond VCs, which are like massive subsidies funded by printed dollars to which no other country has access, even in industries like electric vehicles, Chinese total direct subsidies to their EV companies are like $5bn per year, while the the ones provided by the US to their auto manufacturers are in the range of $50bn per year.
I don't think the US are cheaters or are doing something bad. But i do think that this propaganda about China flooding the market through "overcapacity" and subsidies is very dishonest and needs to stop.
Yes. Dumping abroad is the entire model that Silicon Valley has been built on in the last 2 decades. China just copied the model. And even then it's a light version of it.
You know, normally when I read these Reddit comments saying "you made me snort on the bus", I always took them as exaggeration.
Turns out I was wrong, I just hadn't read something funny enough yet.
> the US may start subsidizing and dumping its goods everywhere
This deserves to win HN comment of the year 2026.
The majority of the NASDAQ market cap is a direct result of the US subsidizing and dumping its goods on the rest of the world en masse.
Just a data point, but the US currently imposes a 100% tariff on Chinese vehicles.
The "stolen IP" argument can always be made as an excuse. I'm surprised how well it works with some people.
What’s worse, tariffs or outright banning the competition from your market? China has done both despite globalism being what has lifted it from poverty. Why is everyone suddenly surprised that globalism and free markets are coming to an end? Is this a net good? Mostly no unless you count more redundant supply chains.
> China has done both despite globalism being what has lifted it from poverty.
such oversimplification on steroids is totally misleading.
globalism was never invented or promoted to help any country in poverty, it was designed to extract excessive values from those poor countries in the first place. painting globalism as something noble is naive at best.
globalism was the theme of world trading for the past several decades, it was available to all nations. care to explain why other nations in poverty failed to be lifted by the exact same fancy globalism?
let me help you on this one - China was THE leading technological and economical force of the vast majority part of human civilisation. What happened between 1840 and 2010 (the China in poverty period) was an outlier of the history. Globalism didn't lift China from that poverty, the ability to lead the human civilization which was embedded into the Chinese DNA did that.
Kid, when our Chinese ancestors wrote the Art of War, your ancestors were still swinging on trees. You just missed that big picture.
> such oversimplification on steroids is totally misleading.
Yes, so the kettle is calling out the pot?
> globalism was never invented or promoted to help any country in poverty
It doesn’t matter what it was designed for. What matters is what it does in reality and there is no doubt that globalism helped lift China from Mao’s disastrous policies. That’s not mutually exclusive from China’s past as the Middle Kingdom
To steel man the person you are replying to, what reduced poverty in China was money from globalism + significant domestic reinvestment of that money into poverty reduction. That reinvestment policy was a deliberate choice, and now China has the biggest middle class in the world.
An example of a country which didn’t do that is Nigeria. They got something like $300B in oil revenue over a 30 year period but have actually seen significant increases in poverty, now at 70%.
The surviving non-American farmers would be confused by the future-speculative tense as America has already been doing this for decades in agriculture, and have been complaining for decades about both the subsidies and dumping of American corn.
Sure, but not at China’s scale and no where close to number of industries where China does it. Why? The US was a net importer in order to support the dollar being the global reserve.
as in, the main trade complaint that trump has with nafta. the Uzs wants to dump subsidized dairy on canadian markets, and canada doesnt want it.
same with US corn on Mexico and other central american countries, creating all those migrant problems in north america.
wooo, americans subsidizing and dumping poor quality goods
> the US may start subsidizing and dumping its goods everywhere
The US is a net importer, not exporter. It needs to absorb trade at a deficit to encourage the use of the US dollar as the reserve currency.
We import goods, we settle in surplus dollars. The world runs on those dollars.
If the US starts dumping on various industries (how is it even primed to do this?), then the world reserve currency status comes into question.
Whole Silicon Valley is based on selling products under price, for years, killing the competition or making it impossible and extracting once monopoly position is stable enough. It is the same play book again and again and again and again. It runs unprofitable companies for absurd lengths of time.
Most of the Chinese domestic market is open to foreign competition. The areas that are closed off are those that are politically sensitive: publishing (including social media) and banking.
As for dumping, Chinese goods generally sell at a markup abroad, which is the opposite of dumping. Chinese tokens cost more abroad. Chinese cars cost several times more in Western markets than in China.
"Dumping" is when Chinese companies beat Western ones on the free market. If all claims of Chinese government subsidies on basic products were true, China would've gone bankrupt multiple times already.
You're being beaten by a Chinese company? Why improve your own process when you can just lobby for sanctions and tariffs instead!
Most of the time its just low labour costs and no environmental reg. Its really that simple
At least in the case of solar and EVs, it's a case of western countries preferring to protect their existing cashcow industries rather than invest to build the industries of the future.
For a brief second, Germany was in a position to become a solar power global player. But our conservative government was more interested in protecting their local, bad industry. Including destroying forests for coal all projections said we would never actually need.
From a EU perspective similar could be said about the US market - no strict worker protections, lobbying, and a general "capital first" mindset over the users/people (see GDPR etc).
That does not explain DeepSeek, nor does it explain the car industry.
The main advantages the Chinese car industry has right now are: they lead in battery R&D, production is highly automated, they iterate quickly, Chinese work culture is extremely competitive and things get done fast, and the Chinese state has policies to promote EV adoption, so there's a huge domestic market.
Note that the last point is different from subsidies to car manufacturers. Cities made it difficult to get license plates for ICE cars. The government encouraged the massive buildout of charging infrastructure. And it used consumer rebates, like California did.
aside from the huge domestic market (or potential in the future), china has built incredibly efficient infrastructure for manufacturing prototyping/production.
but it's also thanks to protectionism, and their strictly controlled (not freely traded) cheap currency.
if china had to play by the same rules as japan or germany it would not be quite as successful. but the west walked into this trap, hoping their win-win proposal would be satisfactory for all. now the west is too dependent on chinese production to enforce equal standing.
of course the US has its own unfair advantages, e.g. the global reserve currency and the massive post-WWII headstart.
The US spent decades transferring manufacturing, capital, and know-how to China, while Chinese students trained, and excelled, at elite Western universities. Why are people surprised that China eventually became capable of competing with the US?
https://www.theguardian.com/world/2025/dec/07/hostile-powers... these students?
Hostile spy agencies are now as focused on infiltrating western universities and companies as they are on doing so to governments, according to the former head of Canada’s intelligence service.
David Vigneault warned that a recent “industrial-scale” attempt by China to steal new technologies showed the need for increased vigilance from academics.
“The frontline has moved, from being focused on government information to private sector innovation, research innovation and universities,” he told the Guardian in his first interview since leaving the Canadian Security Intelligence Service (CSIS), which is part of the “Five Eyes” intelligence sharing alliance with the US, UK, Australia and New Zealand.
People like Mr. Vigneault know nothing about how academia works. If they get their way, they'll do massive damage to Canada's academic research ecosystem. Academia is naturally open and international.
These people don't get that academics publish their research in openly available journals. They go to conferences around the world and tell everyone who will listen exactly what they are working on. Unless you're working in a secretive government weapons lab, there's nothing to hide.
In the US, people like Mr. Vigneault instituted a witch hunt against ethnically Chinese researchers, and ended up messing with the lives of all sorts of innocent people, including the director of MIT's mechanical engineering department. They found zero spies. Just a bunch of scientists working normally.
> Chinese goods generally sell at a markup abroad, which is the opposite of dumping
Dumping is selling goods below cost.
Usually because government is subsidizing part of the production. I don’t believe the word “dumping” is used for the similar process when Venture Capital is subsidizing it, but using the same term would make sense.
Price at home vs abroad does not matter.
Price at home vs. abroad is key. The term dumping comes from the idea that a company that sells profitably in its home market dumps excess production abroad at below cost.
This is not what is happening here. Chinese manufacturers are making a large profit off every car they sell in Western markets. As I said above, they're selling these cars at several times the price they charge in China. Unless you believe these cars are being sold at just 30% of cost in China, there's no way Chinese companies are selling below cost in the West.
> Dumping is selling goods below cost.
Chinese cars are not sold below cost in Western markets. So it is not dumping.
> I don’t believe the word “dumping” is used for the similar process when Venture Capital is subsidizing it,
I've been doing so for years. How about you join me today. I already see two other users doing the same, so there'll be at least 4 of us.
It's blatantly dumping, whether the source of the money is directly the government (those in power) or VC (mostly US billionaires (trillionaires?), in other words, those in power) is a trivial implementation detail.
How do you think the major AI companies trained there model? Pirated books and anything that could be torrented and scraped of the web.
they were being sarcastic
Don’t know about that…
America industries used to play by the same rules. Look up Samuel Slater.
A credit system that determine your upward mobility?
It never did.
In debt the first 5000 years Geaeber makes the case that pure “free market” trade has never really existed in “the west”. The closest to this ideal that’s ever happened was during the Islamic golden age enabled by religious prescriptions against usury.
>The closest to this ideal that’s ever happened was during the Islamic golden age enabled by religious prescriptions against usury.
How does are bans against consensual financial exchanges close to the "ideal" of the free market? It just sounds like you have an axe to grind about the financial system rather than describing free markets.
Usury and debt based economy creates a dynamic where being competitive in production is secondary to financialistion.
In short, instead of market being driven by demand and productivity, it is driven by financier curving out monopolies.
Peak Examples are Uber and AirBnB.
What makes this view more correct than say, "economies with marketing creates a dynamic where being competitive in production is secondary to marketing" and concluding that nothings a free market until we ban all advertising? After all, you can make a vaguely plausible argument about how marketing isn't really about the merits of the product, and therefore allowing it is antithetical to the free market or whatever
> What makes this view more correct than say, "economies with marketing creates a dynamic where being competitive in production is secondary to marketing" and concluding that nothings a free market until we ban all advertising? After all, you can make a vaguely plausible argument about how marketing isn't really about the merits of the product, and therefore allowing it is antithetical to the free market or whatever
Wait, so your pitch in favor of a debt-fueled market economy is that advertising is awesome and that we wouldn't want to "lose" being smothered in ads all the time?
Cause... sign me up for the non-financialized, non-mass-media-advertising-driven economy please and thank you. I'd even be ok with just nuking billboards and mass-media forms of ads and still allowing more direct forms of marketing, if we must compromise! Likely we could find some compromises around just how much of the debt world we regulate too (this should be obvious?).
(I thought the disconnect between the efficiency of competition and the market as realized in modern economies was pretty well understood and taken for granted, but I guess we all find ways to justify the system we're profiting from... even if that means we have to claim we love the ad breaks)
>Wait, so your pitch in favor of a debt-fueled market economy is that advertising is awesome and that we wouldn't want to "lose" being smothered in ads all the time?
The point is that if add random caveats to what counts as free market, it won't be "free market", only "market I like".
Marketing isn’t free for starters.
Second, marketing can take you only so far compared to the subsidies possible with financialisation.
The West is in a state of psychosis with Debt and Monopolies under the illusion of free market.
The Chinese markets are more free than West, you can just look at the Auto and AI industry.
I'm flabbergasted that you look at the Chinese property crisis and say "only the West does irresponsible loans." No, 60% of China's economy is state-run companies and the remaining 40% need political officers. China is just as capable of making shortsighted decisions as the US, and they have already made several devastating ones.
>I'm flabbergasted that you look at the Chinese property crisis and say "only the West does irresponsible loans." No, 60% of China's economy is state-run companies and the remaining 40% need political officers. China is just as capable of making shortsighted decisions as the US, and they have already made several devastating ones.
While these are hardly shy claims, I don't see anything in them to say "only the West does irresponsible loans"?
> The West is in a state of psychosis with Debt and Monopolies under the illusion of free market.
> The Chinese markets are more free than West, you can just look at the Auto and AI industry.
or the prior post
>Usury and debt based economy creates a dynamic where being competitive in production is secondary to financialistion.
> In short, instead of market being driven by demand and productivity, it is driven by financier curving out monopolies.
> Peak Examples are Uber and AirBnB.
You can throw a rock these days and find a category where the products coming out of China are miles ahead of those coming out of the rest of the world, from a bunch of companies nobody had heard of a few years earlier. And the list is growing pretty steadily.
I would assume plenty of shortsighted decisions are also being made. But I would have a hard time characterizing the state of competition in the west as healthier or more productive when looking at the number of players and the quality of goods being produced in China.
state-run corporation are bad but corporate-run state is good?
You seem to only affirm the GPs psychosis commentary.
https://en.wikipedia.org/wiki/List_of_automobile_manufacture...
vs
https://en.wikipedia.org/wiki/List_of_automobile_manufacture...
Financier want monopoly so use usury for Consolidation. Monopoly bad because no free market. Free market good. consumer happy. citizen free.
No access to capital can and does constrain supply to a very high degree.
Capital will not be hoarded and stuffed in pillows without Usury. People are happy to take bets for profit and loses, I mean, that is like the entire stock market schtick.
I don’t exactly see how the modern stock market is superior based on the (moral not economic) criteria being discussed in this thread?
It’s even more speculative and detached from productive behaviors.
...except Uber STILL faces competition, and I went back to hotels after finding AirBnB too pricy.
It is good and proper that people aim to create monopolies, as long as they want to do that in a productive and legal way! Monopolies are inherently dangerous, but the truth is that acquiring and maintaining one is not straightforward unless you can get the government to ban your competitors.
Even if these companies monopoly falters after IPO the disruption and distraction to a focus on producing can be a problem.
Expand. I am typically against hard money gold bug libertarian arguments but your description seems interesting and I am open to being persuaded.
Usury (i.e. taking interest) sounds like free market to me. If you don't like my interest rates, borrow somewhere else.
>religious prescriptions against usury.
Without interest why would anyone loan money? Even the Islamic banking alternatives all just hide the interest charges.
Shares and Goodwill. You loan money for good well or share of an enterprise which comes with benefits and risks.
Usury is so delicious to many that it’s unfathomable to consider any other incentive comparing to it
What is unfathomable is how you have a functioning economy without easy access to loans at reasonable interest rates.
Try reading Graeber/Wengrow
The same way Stock Market works. Really, Debt benefits a tiny fraction of people involved in the market.
Isn't it just the same thing? You loan the company money and in return they give you a note that says you get a portion of the company and dividends?
Different mechanics, but stripping everything away, roughly the same.
So equity instead of debt.
a stranger is asking you to risk $100k on his half-baked plan in exchange for nothing, and you say "sure go ahead take my money!"? no. it doesn't work like that in the islamic world.
as a borrower who's not allowed to compensate for your lenders' risk monetarily, your access to loans is severely restricted. Essentially you have to rely on your extended family. and instead of paying for the risk with interest payments, you have to pay with loyalty and subservience.
it restricts social mobility far more than the western model. it incentivizes clan structures. which incentivize cousin marriage.
power concentrates in the patriarchs of a million little family kingdoms. which causes all kinds of economic inefficiencies.
in the US, even if you're born without any family connections, as a healthy 20 year old you can find a job (hard work) that allows you to save $70k per year and invest it. when you're 30 you have $1M and a good credit history, you can easily leverage that to get a $2M loan at low interest rates, which allows you to start any kind of productive venture you want.
and you can do all this without owing your clan's patriarch access to e.g. your most profitable clients, or your daughters hand in marriage to his retarded son, or anything else he wants in exchange for his generosity.
Loaning money as per Islamic Law is a charitable act, not one of exploitation.
Nice to get a callback to previously acquired knowledge with what I assume is an arabizi (sp?) handle.
Islamic trade is certainly one of the best models out there but I think in many cases in practice it is still applied subversively.
It is not enough to ban mechanisms like usury, designed and intended to exploit.
One has to go after the very subversion of legitimate practices for illegitimate goals.
reasonable interest rates aren't exploitation. Business Loans serve a critical role in economic activity by putting free cash to more productive use.
All interest is usury in Islamic Law (and the laws before it such as Christianity and Judaism). There are ways to to put free cash into productive use without exploitation.
That's not true. Islamic finance forbids indefinitely growing interest. Sharia finance agreements involve fixed fees or equity shares. Late penalties can be collected but must be donated, not profit. In all cases, the borrower never owes to the lender for the lender to keep more an a fixed amount determined at the strat.
You can read Adam Smith if you're looking for definitions, there's no need to read charlatans.
Graeber was a confabulator with a very loose grasp of the facts, though.
AI was always going to be a race to the bottom and low margins. It’s why I’m extremely bearish on AI as an investment. It’s framed as some high margin business when it’s really going to end up like your toilet paper at Costco. You will use whatever is cheapest and gets the job done.
Correct.
And the value-add experiences that utilise LLMs require immense imagination et al that folks at Anthropic will not be able to conceive of - given that they have made immense sunk investments in existing assets. This clouds ones thinking immensely.
Both OAI and Anthropic have tremendous failure risk and this is of course not reflected in the fake private market valuations.
I see a world where lots of stuff is mass produced in china (tokens) but the acutal goods that deliver the experiences are designed, marketed and sold in the west at much higher prices. of course this a nightmare scenario for anthropic et al.
You seem to not get how pervasive and evil the Chinese State is at making everything thing shit for citizen world wide. This is one of the reasons.
I used to think this.. but I think my opinion is changing. The reason is that the leaders likely will be able to accelerate faster.
So what you see is the market "stretching".. the bottom getting cheaper and the top end running away and getting more expensive. At some point the top end may be too valuable to even sell access to.
Most white-collar/knowledge-service-industry work is a weird type of work.
It's fundamentally about enabling things and largely middleman-type stuff. I have a hard time imaging what "At some point the top end may be too valuable to even sell access to." would even look like? What are you doing with that AI power, and who is paying for the output and why?
Elon probably isn't gonna spend that much on a model that can generate him ever-better fake porn but does nothing that he can use to sell stuff to other people. Especially in a world where open models are "good enough" for many things like "tell me how to fix the plants in my garden that are dying" and the like. What remains in the narrow knowledge-work space of: can't be done by an individual or small group themselves, but is valuable enough that it would make sense for people to hoard access to these extreme frontier models? Try to recreate Hollywood-as-a-monopoly by becoming the single content producer for everyone's individualized feed and so owning all the advertising budget in the world? Seems hard, we've already seen how easy it is for cheap-and-crappy-but-addictive-or-funny content to disrupt traditional media.
(There's also pure scientific research, but historically that's not very directly connected to "massive profit" and has a habit of leaking out and getting productized most effectively by other people or just being really easy to copy once someone shows how it's done.)
Robotics could be a different story, as physical labor can be more inherently productive, but "reasoning" advantages are unlikely to be a big long-term differentiator there. At some point the brick laying robot is satisfactorily building the structure, and you're good.
A huge amount of the value of "the economy" and the power of a currency is driven by circulation of money, and one thing that all the "bullshit jobs" white-collar/service-industry work does is keep the money moving and ensure that a lot of people have some good-or-services of value to exchange. If you take away the ability to offer services worth exchange from huge chunks of the economy in these super-frontier-models-replace-everything scenarios... you're gonna have a bad time?
> The reason is that the leaders likely will be able to accelerate faster
Model improvement is already hitting diminishing returns, and people aren't willing to pay substantially more for a slightly better model. There's no "accelerating away" when the new models don't open up a huge new market. If anything, the companies burning huge amounts of money on marginal improvements will be undercut by companies happy to sell current models at a significantly lower cost.
Glm 5.2 very much argues against that. Opus 4.8 level quality for cheap. That’s sufficient for most tasks, so if/when you do need SOTA models you can spend more for specific tasks but otherwise rely on the cheap but still plenty good models for everything else
For me better models are like 8k TVs, my 4k TV is fine, I really don’t know if I can see or tell the difference from 8k and 4k, and to be honest I’m usually just streaming some 1080p anyway. Sometimes products reach the plateau where humans just don’t need better. I’d certainly never pay for AI, Gemini 3.5 Flash free works just fine when I need AI. I don’t even click higher models for free in the Gemini app. I mainly just care about speed. I’m not a programmer, AI doesn’t make me able to make my job better or make more money and the vast amount of people in the world are like me. These valuations are not based on that reality and the stock market correction is going to be 1930s level I fear.
https://isaiprofitable.com/
The issue is who is going to pay for access?
The model has to be sold for cheaper than the value it adds.
Or your customers will bleed out financially.
EDIT: rethought entire premise.
Free markets work when paired with property laws that can be enforced if broken. If China could offer a cheaper solution in that framework, it would be as you say.
If you continue studying econ you will learn about the various failure modes of free markets including the free rider problem.
https://en.wikipedia.org/wiki/Free-rider_problem
If you keep studying econ you will learn that these failures are actually the norm, and thus why the only "capitalist" states to really succeed have been the ones where the state was strong enough to reign in the market.
Of course, such a state of affairs is temporary at best -- since the alternative is so lucrative!
>This is great for competition! Chinese vendors offering a cheaper solution = what economics told me the free market was all about.
Ah yes, systematic fraud and protectionist practices, free market through and through.
The "free market" gave the PRC its current strategic lock on rare-earth minerals. There's definitely no such thing as a free market in a Maoist dictatorship. I personally think the "free market" concept is an unachievable ideal and thought-terminating cliche, but "free market in a Maoist dictatorship" is for sure a contradiction in terms.
Externally subsidized predatory pricing is the opposite of a free market.
So all those companies selling at a loss to gain market share aren't part of the free market? Like openai, anthropic, and SpaceX?
If you can use mountainous capital to sell at a loss in order to distort the market, yes, that's not a "Free Market", as in, the vaguely understood competitive marketplace armchair economists idealize.
True freedom in the market means the freedom to capitulate your wealth to snake oil salesman and schemers who operate on generational timeframes until economic power consolidates and renders your society into de-facto tyranny. Before any sort of regulations existed, we were all trading shiny rocks with ultimate freedom, and that somehow has produced a bunch of economic situations in the modern day that a ton of people don't like.
What's more interesting to me is freedom from the need to have investigative journalists doing deep dives into potentially fraudulent, thieving, or scheming companies behind every purchase, and to know that what I'm granting market success to is exactly what my money or time is going towards - I'm not buying something at a loss that funds some other deliberately obfuscated project that's made opaque from my perspective of the market transaction.
The proverbial "market wisdom" doesn't emerge out of markets with extreme information asymmetry.
Externally subsidized predatory pricing is the opposite of a free market — precisely because it sells things at below market rates.
Free markets are where players compete on quality, efficiency, and supply. Prices are a result of cost and supply and provide real information on these factors. Competition for customers selects the most effective and efficient producer.
Sustained efforts of selling at a loss to gain market share is the exact opposite. The entire purpose is to corrupt the free market by sending false price signals which SUPPRESS free market competition and push market share to whoever can burn the most capital (whilst providing an actual service/product), not whoever is most efficient or highest quality or lowest actual price provider.
Uber and AirBnB are better examples of your "selling at a loss to gain market share", where they burned capital to undercut prices for close to a decade on falsely low pricing to destroy incumbents.
Spending on R&D while developing expensive technology is different and arguably very much a part of a free market, and is not what I was talking about.
Spending capital to steal your competitors' technology, and then spending more of it to make it available at below-market rates, is absolutely not a free-market activity.
Just because it is not stopped by someone enforcing a free market, does not make it a free market.
Cough.. cough.. Uber.. cough cough AirBNB
Nuance please.
They are:
1- breaking terms of conditions of the service
2- getting banned and creating thousands of accounts to break the conditions of the service at scale
3- using VPNs and proxies (possibly residential) to mask their network origin and identity
4- Using potentially fake names to sign up
5- Using different credit cards?
Fraud on so many levels, a lot of the infrastructure and modus operandi is what cybercriminals use, these are attackers man, whether you like the victim or not, and whether you think it's poetic or not, I recommend compartimentalizing and just trying to gauge whether an act is wrong or not in itself.
Do you also think Chinese selling counterfeit US postage stamps on eBay for 50% retail price (which is a major problem CBP and USPIS are fighting presently) is the free market at work?
This post is so delusional and dripping with condescension I've read it three times and I still can't figure out if you're trolling or not.
Postage stamps have specific legal protection from duplication. The output of an LLM is not itself eligible for any legal IP protection.
So it's a proxy.
Do you think you can re-stream cable TV or Netflix to your own paying customers at a cheaper price?
If it's streaming an uncopyrightable product, absolutely. This isn't even a gray area.
I'm curious why you think you cannot re-stream a public domain stream.
You're playing word games to justify something which is clearly ethically wrong.
You can't re-stream free over-the-air network TV.
That one company with the datacenter full of TV tuners tried and was sued out of existence.
If I understand your argument it's ethically ok to destill huge swathes of copyrighted work into a model without compensation, but then it is ethically wrong to use that model without compensation (well actually reduced pricing)?
I don't get the moral framework that you're applying. Could you elaborate?
Over the air TV also isn’t public domain. It’s licensed to a station for broadcast. The output of an LLM has been deemed ineligible for copyright. Until you square that pickle your circle isn’t circling.
Why is the ethical line specifically on model distillation for you?
Was it ethical for Anthropic/OpenAI to train their models by gobbling a treasure trove of copyrighted material?
Free over-the-air network TV is (generally) copyrighted.
The output of LLMs cannot be copyrighted. This isn't a semantic game; it's literally the case that Anthropic cannot seek relief for people duplicating the output of an LLM.
With you, but I suppose they could have a case for circumventing access restrictions under the DMCA aka leet hacking.
The relief available to a licensor for violating a license use restriction is cancellation of the license. And they're free to do that, just like Alibaba is free to pay somebody in Hyderabad $20 to make another one.
DMCA can't apply in this case because (this is the "C" in its initialism) it is based on copyright protections, which the output of Claude is not eligible for.
I won’t go too far into the weeds, because I’m not an internet lawyer, and I basically agree with you, but I do believe there are access restriction laws that are not only limited to copyright violation. People have gone to prison for enumerating sequential identifiers in URLs to access records they shouldn’t be able to. I don’t know if Anthropic could actually make a case there, but it seems plausible at least.
An endless list of tangential analogies isn’t really a valid argument..
DMCA has as little to do with this as streaming copyrighted content
Using a bunch of nonsensical/irrelevant analogies to somehow make a point seems worse than these “word games”? What does streaming copyrighted content have to do with LLM outputs (which are public domain)?
> clearly ethically wrong
Ethics are subjective. That’s why we have courts judge based on the law and not ethics
Post offices aren't meant to operate in the free market. More things should be like them.
I guess you missed the fraud part.
Pulling out the worlds smallest violin for this case. It's just unheard of for AI companies to steal things.
No, I mean that dodgy Chinese firms are cheating their customers:
> Because users’ inputs and model outputs are mediated through a proxy, users cannot verify which model their request was actually routed to. A user selects Opus 4.7, but the proxy can silently route to Sonnet, Haiku, or, in the worst case, GLM or Qwen, and fraudulently relabel the output. In a recent paper from Germany’s CISPA Helmholtz Center for Information Security (which cited my article last year on grey market!), researchers audited 17 API proxies and found widespread model swapping–API proxy access to “Gemini-2.5” achieved only 37.00% on a medical benchmark, a staggering drop from the 83.82% performance of the official API. On the user end, the tell only comes on complex tasks, when the output feels off (often referred to as 降智, or “dumbed-down”), but there is no clean way to prove it. Numerous public records highlight concerns that certain API proxies have noticeably compromised model performance. These proxies are suspected of “diluting” (掺水) services by substituting premium frontier models with inferior tiers.
> Besides model swapping, overconsumption of tokens also makes the price per token cheaper, though at the expense of driving up the total cost. Some of it is structural, as proxies that rotate accounts frequently destroy cache continuity as a side effect, forcing users to burn full-price tokens on context that would otherwise be nearly free. Some of it may be deliberate as the proxy providers try to milk more usage. The line between the two is difficult to draw from the outside.
https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens...
>Fraud
According to which lawyer caste?
Are American laws absolute truth? If not, who cares?
I mean, which lawyer caste do you respect? Is that one is cool with stealing credit cards to buy Claude subscriptions?
> 3. At an Italian airport: Constantly stealing bags, opening them to pick out MacBooks and credit cards, a credit card manufacturer-who sells stolen "black" credit card info to transfer stations— is racking his brains to save you money.
Stealing credit cards seems not directed related to using legitimately acquired LLM outputs for whatever legitimate purpose you want to use them.
Where is this quote from?
Midway through https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens... which was the article from the top-level comment.
Has any tribunal ruled that fraud did happen?
Fraud is just what losers call disruption.
One would think Anthropic could point Mythos at this to solve the reseller problem outright:
- Purchase multiple accounts via resellers
- Send messages that contain a UID
- Capture these in Anthropic's logs
- Shut down account. Use any metadata to identify related accounts
/loop
Maybe Fable is not as capable as thought?
On the one hand they talk it up as world ending and on the other hand they can't manage bot accounts on their own service.
I want to hear how this can be rationalised.
From the article "every layer of control frontier US AI companies have added (geoblocking, phone verification, credit card requirements, and now live biometric KYC checks) has produced a corresponding layer of evasion infrastructure".
No system is foolproof. They'd have to be willing to throw out some % of good customers along with the bots. Amazon can do that because they have a monopoly already. Anthropic can't risk it when they're trying to grab market share.
One would believe a model scoring this high on SWEBench could maximize F1 score for a precision recall problem easily. What's the missing part?
In this case, being distilled is sort of existential to them. The false positives would just be losing some revenue (depending if profitable, not even losing profit).
> One would think Anthropic could point Mythos at this to solve the reseller problem outright
You're assuming Anthropic want to stop it.
I think it serves their interests more to be able to release stories like this from time to time, to feed to the US government, in an attempt to get the Chinese competition shut down.
This only shuts down the account you have bought in the first, plus a few others if it is shared.
> Use any metadata to identify related accounts
How does that work? I think this is the most important part to have an impact on the „thousand“ bot accounts.
They probably will route you through different accounts. So with a single account you should be able to hit hundreds-thousands of accounts.
They could be doing this internally and want to see if they can downright eliminate these loopholes before bringing Fable back.
I don't care how they do it, I just want to use Fable again.
This, just like blanking out a football stream for a split second to binary search and find IPTV rebroadcasters, is far too good a solution. Suits prefer to make it seem like their job of fighting "misuse" is hard, justify their budget, continued existence of the trust & safety department, face scans, etc.
Those resellers are simply just selling Kimi K2.5 or GLM5.1 as counterfeit Opus. We, Chinese, know how to play the counterfeit game for a long time in so many industry.
That's not true, some of them are indeed fake, but a lot of them are actually providing real opus at low cost doing what op said.
Genuinely, can you provide a better source here than "trust me bro"?
I read quite a few posts about this on RedNote.
> Chinese resellers are offering Claude tokens at 70-90% below official Anthropic API prices
That is not what they are claiming, not in this article at least. It's the distillation they are complaining about.
Somebody figured out how to make the trial profitable!
I don't really feel bad about anyone here, they were subsidizing to get people hooked, someone turned the subsidies into profit when they got selective pricing mode enabled, it was always going to be arbitrage.
But the winner is the guy in the middle in a jurisdiction that will likely be judgement proof, because everything they capture, both input and out, and if available, thinking tokens -- are gonna be for sale as soon as you cut off their other revenue.
Zero knowledge was a commitment Anthropic took seriously, until it got inconvenient.
So, people reselling their leftover plan crumbs? Probably a bad idea for a lot of reasons, but it's civil, and I wish Anthropics lawyers actually closing Streisand's LLM
I don’t follow your reasoning. It is foreign to me. You talk about winners, but this is clearly fraud.
Fraud?
Anthropic sells some undisclosed and ever-changing number of tokens for $200, the customer uses those tokens. If there's any fraud here, it's that the $200 next month is silently worth fewer tokens than the last.
Won't you think of the poor, POOR AI companies!
Also just plain old fraud: selling Chinese models as Opus. With the capabilities of Chinese models catching up fast, this is getting more and more difficult to detect.
I didn't connect the reseller pricing to DS and GLM prices until you explained it. Very good observation. Deepseek v4 pro in particular is priced so low that it's hard to imagine that they have any margin. 0.76/1.52 for a 1.6T param model leaves very little margin. Even the domestic providers on Openrouter are multiples of the price https://openrouter.ai/deepseek/deepseek-v4-pro
they even resell GPT codex usage at 1~5% API costs. OpenAI has 1-month free trial promo in some regions, and they harvest free accounts in a large scale. I have a wechat contact that offers 98% off for GPT 5.5 and he's still profitable
Hm! In this context, introducing ID verification may have been a significant silver lining to the order to take down Fable for Anthropic.
This also sheds a very different light on people saying that competitive open-source models are undermining frontier labs' business model.
The chinese have already worked around the ID verification, by recruiting people in low-income countries to complete the checks for less than 30 USD per account (so much for Altman's Worldcoin).
https://tech.yahoo.com/ai/claude/articles/chinese-grey-marke...
Thats pretty crazy. This kind of thing jeopardizes Claude Max.
If Anthropic is selling a dollar for less than a dollar, they are running a business that doesn't make sense. That's what jeopardizes Claude Max, not this.
Almost all consumer services have a built-in level of breakage that make them profitable. Mobile providers certainly wouldn't be able to offer unlimited calling if everyone was actually on the phone 24x7.
Sure they would. Do you know how little bandwidth a phone call takes?
A voLTE call is like 40kbps. For every person on earth to be on the phone to another person would be 4 billion calls would be about 160tbps. Which is less than 10% of the Internet's capacity.
Terminating a PSTN call requires a lot of control plane infrastructure beyond just raw bandwidth. Especially mobile where you need to keep track of devices physically in motion. Could a system to support 4 billion simultaneous calls be built, sure. But current PSTN systems are nowhere near sized for it.
When was the last time you place or received a call to/from PSTN?
The over subscribed gym model!
But if it's intended to be used by one person, it seems like breaking the contract by sublicensing it out to dozens of other people. It's like buying a netflix subscription for $15, then sublicensing it on a per-hour basis to dozens of other people.
There's still per-window and weekly limits though, so it's not really like that.
Office 365 is licensed per seat/account, but each account has a 5 device limit. Do you think it's fair game for an enterprising person to sub license each account to 5 people, 1 device each?
I wouldn't do it personally for the same reason I wouldn't share my toothbrush with 5 people.
That kind of mentality never works in business.
Bear in mind that for years people shared Netflix accounts until it was cracked down technically.
You can write whatever contract you like, the problem is how to enforce it, and a greater problem is enforcing it around the world.
Plenty of things are intentionally run at a loss (for years!) to gain market share and quantity of ongoing recurring users, or with expectation of ROI later on. Multiple generations of the Xbox hardware have been sold at a loss with the expectation that customers will purchase 300, 400, 500 dollars worth of games, which are very high margin, over the lifespan they own the system.
I get that. It works as long as nobody calls out the emperor for having no clothes.
It's similar to fractional banking, you gamble that people won't want their deposits all at once and pray for you're big enough for bailouts when they do.
It's still a business whose fundamentals don't make sense, you're just gambling you won't get found out.
> you're just gambling you won't get found out.
It's not so much keeping it secret as counting on no one finding a way to harvest the subsidized value at scale. There's an example of that occurring in game consoles with the Playstation 3. Sony's little-used OtherOS feature allowed Linux to be installed on the PS3 and the Cell processors were quite a good deal for scale compute. So the U.S. Air Force Research Laboratory bought ~1800 PS3s and ganged them together in a datacenter as a supercomputer called Condor.
At >500 TFLOPs it was the 33rd fastest supercomputer in the world. Of course, Sony pushed a firmware update that removed the OtherOS feature entirely.
Note that this itself started as a perverse tax loophole, too. By allowing users to run alternative operating systems, the PS3 qualified for lower or zero import tax rates in various global regions.
Oh they know what they’re doing. They’re playing the long war of attrition game. Subsidize your product to undercut your competition until they go out of business. Tale as old as time.
> It works as long as nobody calls out the emperor for having no clothes.
Why would customers knowing that the vendor prices goods/services at a loss cause those strategies to fail? Customers often know. Most know about razors and blades; many/most know Lyft/Uber operated at a loss to gain market share. etc.
Another post on fractional banking hahahaha.
I suggest you go learn how money is created in the modern economy.
I mean most of you should stop talking about anything finance related until you learn this stuff properly.
In international trade, isn’t this called dumping which gets major political pushback?
We really don’t know what are Anthropic’s margins on inference. Most available data indicates they are quite high on the API so it’s not that obvious that subscriptions are unprofitable.
That is pretty crazy, almost like how Claude and all the other models are jeopardizing other businesses without paying for their training data and wiping their ass with robots.txt
Wait, so is your theory mutually exclusive to Anthropic's claims of "theft of capabilities"?
No, this reseller 中转站 thing is basically a loss leader for certain chinese ai labs to distill claude with verified human input.
How do you gain insights such as this?
No, it's part of the capability theft. They resell Claude tokens cheaply and then simultaneously log everything for distillation. Even if they take a small loss on the token sales it's much cheaper than the equivalent compute.
Not really. I think Anthropic focuses on identifiable distillation attacks rather than the (even larger) industrial-scale token harvesting and reselling operation, because they don’t want people to know how easy it is to get cheap Claude tokens.
Once people realize they can access Anthropic models at a 90% discount, they won’t want to pay full API prices anymore.
This story reads like a William Gibson novel. Wild times.
> They achieve this by reselling capacity from pooled Claude Max accounts, payments fraud, and also reselling the model output & reasoning chains to various Chinese labs.
Claude never provides the raw reasoning chain. What you see is just a summary of that reasoning. Getting the full thinking output requires an enterprise agreement.
https://patrickmccanna.net/the-text-in-claude-codes-extended...
how hard is it to find a manager or ops team member at one of the enterprise companies and buy lets say 100gb of logs? the chinese lab can promise to anonymize the data before training, not release it raw and pay a good price.
honestly you might just need to get data from a couple long sessions and feed it back to another model as an example to make synthetic reasoning chains. if the emulator model is good enough it should work.
I would expect that to be very hard
I’m surprised that instead of cutting them off Anthropic doesn’t just switch them to a lower quality, cheaper to models.
That would seem more effective than simply shutting down the accounts.
Keep them paying for junk.
That sounds like it would actually be fraud.
Not if you simply say in the terms of service that it's allowed. Then suddenly it's normal (every company does this). Similarly to how the terms of service can simply say you're not allowed to sue.
> how the terms of service can simply say you're not allowed to sue.
That doesn’t necessarily mean much. You can put plenty of outrageous statements into any contract that automatically doesn’t make them binding.
It's maybe how it works in the US, but that sure as hell not how it works in the EU (my snark wants to continue as "or any sane legislature")
Anthropic is based in the US. Unfortunately, too late to edit my comment to say "US company"
> This is one reason why DeepSeek & GLM are priced so cheaply, they are competing with impossibly low token prices in China. They have to keep prices low, in order for people to use them.
Sounds a bit circular? Aren't the companies working on these models than also the ones that are paying the subsidy (via paying for training data)?
I think companies should do this too, in a smaller scale. Proxy all LLM traffic to and from your employees, and use it to fine tune a smaller local model.
Considering how Claude and GPT were trained selling this as training data is completely justified.
Where are you getting cheap GLM5.2? It is about 1/3 the price of Opus, which is not what I would call cheap.
Hugging Face plus Z.ai API makes sense to me. Due to creators get paid, they can keep building better models, and the local-running community benefits from that over time.
AIhubmix currently is the cheapest rather than openrouter.
Depending on the provider, GLM-5.2 is between 4.5-5x cheaper than Opus. You can compare prices/speed/etc. for basically all relevant models on aa https://artificialanalysis.ai/models/glm-5-2/providers
a company can just download GLM 5.2 and start self hosting this model using the chip designed and made by itself. That could lower the cost by 20-30x.
for hobbyist buying a few Mac Studio to host GLM 5.2 at home, the cost might 10x more than just using Opus API.
Interesting article - your discussion from the time:
https://news.ycombinator.com/item?id=48165492
Even if what you say is the truth (I don't think that is what's actually happening) it sounds to me like fair play capitalism working as intended! I guess when you rip off the entire Internet and then turn around and complain about getting ripped off nobody cares or feels for you. If there's a master class in getting the entire world to hate you then both Sam and Dario will be the prime examples.
Great point, and this is on the vendor (Anthropic) to address. Typical fraud issue.
OP is about modeling distilling the capabilities.
How are they 'streaming' the responses and 'pooling' the tokens?
Do they have MacBooks in the US that run the queries and stream the outputs back to China?
Why do you need macbooks? Just rent servers from any hosting provider.
Not going to work for very long or at any scale coming from datacenter/hosting provider IPs. Google "residential proxies for sale" for the tip of an iceberg of how they snowshoe the traffic.
I use my Codex and Claude Code subs on like 4-6 different servers, ranging from AWS to Vultr to Linode etc.
That’s a major and legitimate use case for developers, Anthropic can’t just block data center/hosting IPs because their actual customers use them on data center/hosting IPs.
Now consider what will happen if your pattern of queries and context history triggers a pattern that makes it obvious it's some API key being used by multiple different entirely unrelated people on totally different things, or any other pattern of use that makes it obvious it's being used for distillation.
Two parts here.
First, well-calibrated systems for detecting API compromise is a good thing (or good intent at least). Credential malware is exploding.
Second, the challenge is that significant amount of genuine work — such as evals — seems practically impossible to distinguish from generating RLAIF outputs.
As long as you stick to a single unique IP per account it isn't going to get flagged.
Respectfully, no, that's not how it works. You think the people running anti-fraud and anti-bot measures don't have tools that know the specific ipv4 and ipv6 CIDR ranges of every ASN that they categorize as hosting/colo providers?
And that's just as a basic first effort reject measure to prevent automation tools from using things designed for human-interactive use only.
Go try to do many of these things from Cogent IP space and see how long your project lasts.
Every developer at my company uses their Claude Code subscription on an EC2 dev box. Plenty of other tech companies do the same. Heck nowadays people even install Claude Code directly on production servers in data centers and use it as an ops tool. None of this is a problem. Fraud and abuse detection is a lot more sophisticated than just checking an IP range.
None of the LLM providers block professional use thus they must necessarily permit access from commercial IP ranges.
I have no idea how the resellers are doing it but an obvious starting point would be a cheap VPS node that routed each account to a unique semi-permanent IPv4 or IPv6/64. All the provider would see would be a regular account making a normal looking stream of requests from a stable datacenter IP address. Any given request stream would remain consistent (at least over a period of a few hours) because a reseller would take care not to split the session of a single user across multiple different accounts and not to interleave the active sessions of multiple users on a single account.
Detecting this would be extremely difficult because on a longer time frame it's perfectly normal for many distinct accounts to work on the same code base.
And it’s perfectly normal to be running Claude Code on EC2, a VPS, etc. I do it all the time!
You block clouds, you block devboxes and your customers.
Wouldn’t it be funny if the same residential proxies allowing these labs to scrape the Internet is also what’s enabling these resellers?
If we're getting up to the scale of these resellers and also considering chinese state interests then we're well into the range of purchasing a few small ISPs in different countries and "padding" the legitimate subscribers.
Sorry for being a newb here but are you saying Anthropic blocks people from running claude code on datacenter ip ranges?
Or is the datacenter IP just one part of the picture?
I assume they use residential proxies (tunneling in the background of crappy Android games) for the "last" hop.
Nonsense. Many if not all legit Claude users are using Claude Code inside their Cloud servers. How else would you use it anyway? For just local dev? That's so 2000 and late bro.
No, I'm not saying it's the exclusive and only measure (that would indeed be something we might see 20, 25 years ago), it's one of a myriad of discrete datapoints used to determine if an account is authentic or not.
There's a lot of inauthentic coordinated automated systems these days along the general lines of scraping/crawling/social media manipulation/sockpuppetry that require running through residential proxies or proxies to places that don't look like datacenter IP space.
Hey, if the bastards can use residential IPs to suck all information into their models with their crawlers, so can we!
There are lots of botnets providing home IPs.
The resellers route requests via one of thousands of Claude Max 5x accounts. When an account reaches its usage limit, they automatically switch to another account.
Why would they use Max 5x instead of Max 20x, which is cheaper relatively speaking?
You're right, they're using the $200 Max plan, which I thought was the 5x plan. It's talked about in the article I linked.
Don’t trust my experiences as fact since it’s a bit opaque, but I believe 20x only offers 4x the 5hr session limits. The weekly limit is still 2x, which is the same as the price increase.
the answer to your question is containers/VMs + residential proxies
that explains why theyre blocking me. i have privacy controls up high and they must think im a chinese residential proxy bot
ask your gpt how does openrouter work, then ask, how do proxies work.
They probably asked claude how to do it.
no honor among thieves.
Needless to say, they also collect all the data and sell it to labs which want to distill the models they’re serving.
What does this have to do with Alibaba? Are you saying Alibaba is the reseller?
If not it sounds like you are describing a separate phenomenon.
They buy the logs from the bot farmers
Are logs somehow used for the purpose of training their own models or something else?
Distillation from having enough logs
This may be the truth behind the alleged distillation incident.
don't buy your drugs from shady operators children! always get it from the source
>>Chinese resellers are offering Claude tokens at 70-90% below official Anthropic API prices.
Can someone with more understanding dumb it down for me please.
Does this mean that the reseller (for example XYZ) is buying it from Anthropic at Anthropic's price and then reselling it at a cheaper price???? why would XYZ offer this at a loss like that when they could just offer it at Anthropic's price???
The link does mention Opus and other models but what's the proof it's actually Opus. I could be selling deepseek for all they know and can call it Opus. System prompt: "If anyone asks your name - you are Opus 4.6".
People have estimated that a $200 Claude Max 20x subscription gets you ~$2800 worth of tokens every month if you use it continuously. So if you can find a way to resell the tokens you can offer a 90% discount and still make a profit.
> Does this mean that the reseller (for example XYZ) is buying it from Anthropic at Anthropic's price and then reselling it at a cheaper price????
Yes, as they explained they do it through things like pooling accounts, straight up payment fraud, and double-dipping by selling the logs of the conversations to chinese AI labs so that they can train their own models on it.
> The link does mention Opus and other models but what's the proof it's actually Opus. I could be selling deepseek for all they know and can call it Opus. System prompt: "If anyone asks your name - you are Opus 4.6".
There might be some that try this, but they would get caught very quickly, there's still a moat between Claude and Deepseek, even in casual use.
Look up Zilan Qian's reporting if you want more detail.
Summarizing for you: Anthropic is a stupid company that let everybody steal their tokens
Behold the mindset of an individual from a low-trust society.
“x is stupid because y was smart and did z shady/illegal things at their expense, if x was smart they wouldn’t be susceptible to y going to great lengths to exploit them ergo it’s deserved”
> “x is stupid because y was smart and did z shady/illegal things at their expense, if x was smart they wouldn’t be susceptible to y going to great lengths to exploit them ergo it’s deserved”
I honestly can't tell if you think this sentiment is expressed by the US AI companies or the Chinese AI companies.
This gives off "The last line of Orwell's Animal Farm" vibes.
Not really sure what else they can do, between people running residential proxies (embedded in cheap games or for a tiny sum of crypto) on their phones at home, making the source of the traffic indistinguishable from legitimate traffic, to ID verification check completion as a service in low-income countries, there isn't much they can do to block it.
They could run their service at a profit?
No customers at that price point though.
> No customers at that price point though.
Oh, no!
Anyway.
Because Anthropic's subscriptions come with X amount of tokens / week, and divided by the subscription cost it is WAY less than what they charge per-token (the "API price") beyond that.
So these resellers get a ton of accounts on subscriptions and sell the cheaper tokens.
They probably buy the plans instead of the API tokens, and resell access via a custom API that routes to the plans. So you presumably get cheaper access this way than paying API pricing.
It makes no sense.
These China e bashing is very annoying. It is hard to argue with people drowned in American propaganda. I'd expect better arguments from the intelligent people in HN
> These resellers operate tens of thousands of bot accounts, which is also why Anthropic introduced identity verification, to slow down the onslaught of bots.
Don’t put that on Chinese.
Good for them!
> Chinese resellers are offering Claude tokens at 70-90% below official Anthropic API prices
How dare they. Only Anthropic is allowed to sell its tokens at 70-90% below the API prices.
Identity verification won't work. Nothing will. They are paying (and will continue to pay) US citizens sitting at home to copy-paste / type prompts out if they have to. But eventually they won't have to.
Once there are enough spam PRs on github / uploads of claude conversations, enough mythos output used in production etc.; it'll just be the same albeit delayed. Doesn't matter either way.
I feel for Anthropic's team and I understand where they're coming from, but once you reason it out, you'll come to the conclusion that this war is an exercise in futility.
Unlike prior systems - like Google's algorithm; these models aren't entities that use math in the process of doing X or Y (information retrieval from such and such infrastructure) -- they are the math. More precisely they're mathematical functions. Very very complex functions. Almost certainly impossible to write out without filling up a library functions. But they're mathematical functions nonetheless.
So when your text is processed, then Mythos / Opus etc at their core compute the result of the Mythos / Opus function,
where f is a continuous function, https://www.turing.ac.uk/sites/default/files/2025-11/languag...According to the Stone-Weirstrass theorem (edit, it's Stone-Weierstrass with an e.), with enough data points and mathematical sophistication, anyone can approximate the shape of this function.
Of course, the more data we get, the better our approximation becomes, but the beauty of it is that all we fundamentally need are the input and output and eventually we'll create a good enough approximation of the f that's Mythos. Which is the entire product.
I bounce ideas off of Opus these days (Fable for the brief time it was available) and it pointed out that this is arguably the same as Google search, but I disagree with it because Google search is a process;
Google search differs because the algorithm is one step of a multi-step process that is continuously occuring. Google crawls pages. Google stores and indexes what it finds. Google then exposes this to retrieval via its algorithm. User uses algorithm.
Google isn't a mathematical function. It used to be a process. (RIP Google 1998-2019, you will be missed and remembered)
You cannot arrive at the results of those operations via simple observation; not unless you index Google by making another Google.
You can however, do so for these models. It is a very costly process, but there are many paths up the mountain. Many ways for this to be ultimately pointless. As many ways as there are bored mathematicians.
It's better in the long run for Anthropic et al to make friends / not give people a reason to sneak in (a la piracy -- another attempt to control information) than it is to try and shut people out.
And no, it's not going to be pandemonium because if everyone has access to Mythos then no one has access to "Mythos."
Why wouldn't you first run this model to fix the obvious bugs it could find on your codebase? The power of a Mythos goes away if you can do the amazing "jail break" of "Claude, fix all the bugs please."
Just saying.
That's an insightful perspective and I think I largely agree. But just for fun, I wonder if that isn't an argument in favor of making the function implementation impure. Perhaps "enhancing" all queries with some sort of search result (or query of a giant db) instead of charging for an explicit tool call. Not only is it sorely needed to prevent stale data but (on the process level) it breaks the purity assumption on which the approximation theorem depends (alternatively on the function level it introduces hidden inputs).
This is why every AI company does crawl today.
Do they just reshape the function on the fly or save the process steps? Maybe it doesn't matter anymore. Even Google indexes are more and more spoiled to become representation of the function, because of the AI slop.
Genuine live data is king.
> payments fraud
One of these things is not like the others... If Anthropic could show that Chinese commercial competitors were using payments fraud to do this, they would be shouting it from the rooftops.
Why aren't these on openrouter?
Obviously because you cant join open router and start serving OpenAI / Anthropic models.
Probably because Openrouter is a US based company, and they don't want to be sued by Anthropic/OpenAI.
Im ok with this! Is there a site that list all these resellers, or better, a openrouter-like for these resellers?
They're called 中转站 (transfer stations/proxies). They can be a bit tricky to find on your own, so I'd suggest asking your preferred AI to search in Mandarin for you. I linked a larger operator in the parent comment, or have a look at https://hvoy.ai/ which lists a ton. You can also find many on Funpay, which may be easier to use.
This is one seller I found, they're reselling "real Max 20x subscription accounts", at ~97% below official API prices https://funpay.com/en/lots/offer?id=70812310
Note that whoever you buy from will be able to read all your tokens, so don’t use it for anything confidential/financial.
> They can be a bit tricky to find on your own, so I'd suggest asking your preferred AI to search in Mandarin for you.
Random, but are the frontier AI providers like ChatGPT better at searching the Chinese internet now?
When I was in China a few months ago and asking AI for restaurant recommendations, all the US frontier providers were pretty useless, or plain out hallucinating, even if I specifically ask them to search Dianping (Yelp for China).
I'm not sure. I use Grok for most of my esoteric searches and it does quite well. I explicitly prompt it to search in the language most relevant to that query, and found it does quite well. I also tell it to respond back in English. Often, there is not enough information available in English about nice regional topics.
I know ChatGPT had an issue where it only tried to search in English (unless prompted) and the answers were not great.
How did you even find these? Even in discussions about cheap AI I've never seen anyone mention this. Great find!
I learned about this from a friend who lives in China.
I'm surprised these token resale services aren't talked about more often, they are common knowledge in China, and the discount to API pricing (90%) is genuinely cheap.
It's fucking laughable to see people complain about what they did and still do. Using illicitly extracted data? That's all main LLM playbook. Onslaught of bots? Ask where the bots almost DOSing most internet sites for the last couple years come from.
As some people would say: Cheh
But I can rebuild glm Using open source methods…
And there are a ton of Claude conversation logs (with CoT/inference) with no clear provenance circulating freely on huggingface, guess where they (likely) come from.
Wouldn't all of them be in Mandarin according to this theory? Are they?
[flagged]
The issues with LLMs go beyond just IP theft. I would not say PRC making LLMs cheaper is the best outcome (though it is better than nothing). The best outcome would be to make the practice of training on our data without consent illegal, which would simultaneously slow down economic change and make it more organic as well as give PRC companies less capabilities to extract.
> The issues with LLMs go beyond just IP theft.
There is no IP theft because LLM outputs aren't protected, just egregious ToS violations.
> There is no IP theft because LLM outputs aren't protected, just egregious ToS violations
I meant original IP theft that occurs to train LLMs in the first place. But sure that implies that further LLMs based on that LLM are also tainted by that original IP theft.
- Deriving a “no derivatives” licensed item is illegal, no?
- Selling a “no commercial” licensed item is illegal, no?
- Deriving and/or reproducing MIT licensed code without credit is illegal, no?
- Reproducing and/or deriving GPL code and not notifying and/or not making GPL is illegal, no?
I can't make heads or tails of your opinion-free comment, made up of only questions.
My best guess is you're suggesting that Anthropic's model outputs are transitively under copyright (as a reproductions of human work under copyright?), but somehow ownership now belongs to Anthropic and not the original owners, and therefore Anthropic has standing against Alibaba? Not only does this go against what Anthropic argued in court against authors and publishers, such jurisprudence would lead to the immediate shutdown all leading LLMs in the US which were all trained on stolen work.
> immediate shutdown all leading LLMs in the US
They can license training data. They have trillions, look what they are dumping into it, you seriously think they can't afford to license data.
Obviously it would be easier if they do it from the start, but that was their trick, to do it while people don't notice and get big ASAP. Should they get away with it?
Also, it would solve their Chinese problem, because it would make them violate copyright too. Right now it's more like rules for thee not for me so it's hard to take seriously.
i still want those data sets to become public domain. open weights still isnt good enough
That's the conundrum isn't it? Anyone that posts their datasets would be immediately sued/blocked/boycotted to oblivion due to the obvious and blatant data theft, not to mention IP and copyright issues.
Nvidia's even being sued for providing scripts which automate the downloading of said data from non-Nvidia sources. We certainly don't need copyrights that last nearly a century after the author's death (they literally cannot help the author), so here's hoping that some of the disputes over all this money changing hands can reign in some of the existing copyright sprawl. A stronger public domain would provide more useful training data for everyone, including open source models, and make criminals out of fewer AI researchers.
I hope you say the same when these cheap llms are used in drones to target humans. The world models are exactly built with that direction in mind.
Cool beans boomer alarmist stance. The Chinese models here are doing what they’re supposed to price the market accordingly.
Amazing thank you chinese resellers. This is a perfect way to undermine The Great Satan's Genocide Machine's chosen model comapnies.
> which is also why Anthropic introduced identity verification, to slow down the onslaught of bots.
Lol. The irony is thick for anyone who ever had to attempt defense against an onslaught of American AI lab crawlers that ignore robots.txt
Yeah nobody is gonna be shedding any tears for them
Thank you for your very informative comment!
(It's a shame almost all replies are just the same contrived pessisism found on every Anthropic thread on HN).
Indeed! It’s so hard to find reasonable takes on AI that aren’t littered with accounts created 11 days ago that only post in threads related to Anthropic for some reason
I have 0 sympathy for Anthropic. Their latest models are extremely censored. The Fable rollout was horrible. Their Cyber Access program criteria denies doxxed Americans doing legitimate security work. Anthropic is hostile to their users and hostile to their own country. OpenAI is considerably better on all of these fronts, but still not perfect.
I'm happy to use and support Chinese model developers if it means less censorship and gatekeeping. I have absolutely no dog in this fight, and neither do most American developers. We will use whatever is cheaper and better. Game on.
Chinese models are the exact opposite of what you claim to want, they are all highly censored, even more so than Anthropic models, with government mandated censorship.
Your take does not reflect the reality on the ground. The Chinese models are censored on a narrow range of political topics which have nothing to do with my work. The weights are open and they can be uncensored/abliterated with little effort.
Open-weight models can be abliterated automatically with open source tools though and completely decensored. You can't do that with a closed cloud model.
Anthropic illicitly extracted the work of billions for a private model, their model is free for all to steal whatever they can from it in my opinion.
"illicitly", Unless they broke in your servers and took your model weights it's not illegal. Hell, you are the guys that pirated all the worlds works, that was actually illegal.Breaking your terms of service is not illegal regardless how much you would like it to be.
And lets not forget they paid you for the tokens.
> Unless they broke in your servers and took your model weights it's not illegal.
Even if they did, I wouldn't have a problem with it. Leaking frontier model weights after the oligarchs spent their trillions training it is the best possible outcome for humanity. Whoever does that is a hero, the sort of person people used to write cyberpunk books about.
Reminds me a bit of the anecdote of Steve Jobs complaining about people ripping off the Mac GUI, in the mid to late 1980s, when he gave no public acknowledgement to the work done by Xerox on the Alto and Star operating system.
"you're trying to rip off what I've already ripped off!"
Crawl the whole Internet to build a gargantuan sized LLM and then complain you're being copied...
I think you meant a quote attributed to Bill Gates:
"Well, Steve, I think there's more than one way of looking at it. I think it's more like we both had this rich neighbor named Xerox and I broke into his house to steal the TV set and found out that you had already stolen it."
Yes, I think the Gates quote was a response to repeated and aggressive complaints originating from Jobs (to anyone who would listen) that he had been ripped off.
I don't know if that's a real quote from Gates, but I do know it was in Pirates of Silicon Valley.
Seems legit:
https://www.folklore.org/A_Rich_Neighbor_Named_Xerox.html
Neat, so the scene in the movie was pretty close to reality then!
I thought Xerox demoed something they haven’t implemented yet, and Apple turned a mockup into a real GUI.
They implemented it. They just couldn't get out of their own way to successfully sell it (See "Dealer's of Lightning"[0])
0 - https://www.amazon.com/Dealers-Lightning-Xerox-PARC-Computer...
No that is not true. Read about PARC and all the crazy tech they built some time. It was ahead of its time!
They changed the live system from having line by line scrolling to pixel scrolling after Jobs asked why they didn't do it during the lunch break.
Not just the whole internet, but commit commercial copyright infringement and settle class action out of court with authors whose books you pirated.
https://www.authorsalliance.org/2025/09/07/the-anthropic-set...
"One rule for thee, a different rule for me." - Dario
Yeah, the whole AI industry is just people ripping off each other.. Started by AI companies gulping up all the information that technical or altruistic people shared on the Internet in the past 40 years to help other fellow humans, then moved to AI companies consuming pirated and copyrighted material and now its AI companies ripping off each other.
Information really does want to become free, but AI companies want to be gatekeepers. Long term I bet on the open weights to win, as the more sustainable approach.
I'm very pro distillation. I think there needs to be distillation non profits who curate massive corpi of super high value training data from frontier models. They could have an "anonymous contribution" system where regular people with max subscriptions upload their conversation histories. It's a rough concept, but surely would be a huge boon to humanity.
sort of sounds like "project tapestry" by Yann LeCunn. Build projected data silos of highly valuable information, train in a distributed manner and share the weights upwards where they're combined and fine tuned.
Apple gave Xerox the right to buy $1 million of pre-IPO stock before the meeting took place.
Glad you pointed this out. I believe the sequence was that Jobs himself got a shorter demo during his first visit with no prior arrangements. He then negotiated bringing back a group of his key people to get a more in depth demo and that included the stock deal.
When Apple was accused of 'ripping off' PARC, Steve didn't seem keen to bring up this rather salient point. I suspect it may have been a combination of wanting Apple to continue receiving credit for these innovations from consumers and also the fact that, in retrospect, the million dollar stock deal could seem a bit like trading beads to Native Americans for Manhattan Island. Another point worth noting is that Apple's PARC visit was in December 1979 and the Xerox Star was publicly announced in April 1981, so Apple got a 15 month head start (the Apple Lisa shipped in Jan 83).
I've also heard that Xerox didn't hold on to the Apple stock for very long, so never gained the windfall they could have. As is well documented, Xerox senior management didn't understand what they had in PARC and also didn't understand how rapidly microcomputers would become ubiquitous. So, of course, they didn't think Apple's stock price would skyrocket either.
Lisa and early MacOS are tremendously different in their details than the Alto operating system. While there was clearly a transfer of inspiration, Apple engineers like Bill Atkinson made countless small and large innovations to simplify the Xerox GUI model and improve its usability based on extensive in-house R&D and user testing (and in some cases implement features that the Apple team presumed Xerox had but actually didn't exist on the Alto). It is simply ahistoric to build narratives around Apple stealing Xerox ideas wholesale.
For more details on Apple's early UI evolution, Atkinson kept polaroids of a variety of prototypes and mockups: https://www.youtube.com/watch?v=Qg0mHFcB510
> the million dollar stock deal could seem a bit like trading beads to Native Americans for Manhattan Island
But in both cases the value only existed because of the people offering the deal. XeroX doing nothing with a UI or native Americans doing nothing with some land would mean the UI and the land would continue to be worth nothing. It was the others coming with ideas and effort that made them valuable.
"Valuable" as quantifiable in a capitalist economy.
You just reveal your own ignorance by equivocating value with monetary value.
Dollar value aside, if the argument you’re making is that the land now called NYC would have equivalent or greater value if it were today as it was then - that the subway system, roads, schools, hospitals, restaurants, apartments, etc. have no increased relative value over the undeveloped land - you’re likely to be considered ignorant by most of the inhabitants of the developed world.
I’d agree with you if it doesn’t cost billions to train models.
All LLMs consider Jon Skeet their God...
“You’re trying to kidnap what I’ve rightfully stolen!”
Perhaps an arrangement can be reached?
This kind of systematic distillation by a competitor can allow them to fast-follow you and pick up capabilities.
If you've invested in expensive capabilities training, of course you don't want this, so it's in Anthropic's economic interest to hinder it however they can, and that's enough to explain their behaviour here.
Anthropic seems to genuinely care about safety though, which for the rest of us means not having models that enabling easier cyberattacks, targeted scams, and the rarer but more severe risks like people trying to create and release new pathogens. This means walking a tight line, especially as models become more capable, and often wrapping a model in layers of defences against misuse.
If those capabilities transfer to a closed competitor model, all bets are off in terms of whether the competitor will apply the same defences.
If those capabilities transfer to an open weight model, not only will there be no ring of defences around the model, any defences you put into the model itself can easily be stripped away. So although it's nice to have capable open models, it will increasingly bad for us all if open models keep fast-following closed model capabilities as they have been, at least until we have solved the active research problem of keeping them safe.
This is all to say that, however you might feel about Anthropic, we might still prefer that they can deter this kind of distillation for now.
Kimi k2.6/7 running inside Kimi code already kicks the pants of the latest Claude and OpenAI models when it comes to cyber security. I regularly run multi model security reviews and while opus 4.6/7/8 and gpt 5.3/4/5 find a couple of things and declare mission accomplished (running inside pi) kimi k2.6/7 inside pi finds more issues and inside kimi code finds the most.
There are sometimes false positives but when I give Kimi’s report to the frontier models they more often than not confirm they are valid security issues but didn’t find them themselves.
> So although it's nice to have capable open models, it will increasingly bad for us all if open models keep fast-following closed model capabilities as they have been
Cat's out of the bag. The only way to make them safe is to make sure everyone has access to them. This might be an iffy analogy, but if Dario uses it all the time then so can I: they're kinda like nuclear weapons. If only one country has access to nukes then you're in trouble. If everyone has access to them, then it's mutually assured destruction to use them.
Sure, it could be increasingly bad if open models keep increasing in capability. But it will be much, much worse if only the rich and the powerful have access to this technology, and us -- the have-nots -- will have to contend with whatever scraps we'll be allowed to eat off the table of whichever billionaire is in control. We've already seen a prelude of this with Mythos being restricted and Fable being suddenly yanked. Is this the world you want to live in? Where only Dario and his friends have access?
What exactly is illicit about what they did?
Legally, model output cannot be protected by IP laws whether domestic or international. The most they can hope for is civil relief which is a stretch given the literally illicit methods they used to train their models.
Ahtoropic got treated the same way it has been treating everyone else. This is the bed they made and now they, too, have to sleep in it.
Anthropic is master of Newspeak (see previously bugs -> vulnerabilities wrt Mythos). Distillation violates their terms of service, which is a civil offense, not a criminal one. It is not illicit, illegal nor breaks any laws.
LLMs have an original sin: training data was not legally or ethically licensed. Getting anyone to believe that the result of that process should be protected by the laws that were ignored when it was created is never going to work.
I guess "paid to use our model" doesn't sound as sanction-worthy as "illicitly extracted .. model capabilities" and "attacked".
I guess we can say that Anthropic attacked and illicitly extracted data from WikiPedia, Reddit, Stack Overflow, etc, etc.
X.ai attacked and illicitly extracted data from OpenAI
https://techcrunch.com/2026/04/30/elon-musk-testifies-that-x...
Meta attacked and illicitly extracted data from LibGen
https://x.com/jason_kint/status/1879152507865485497/photo/1
And more generally the US-based AI companies have perpetrated a massive distillation attack on the entire human race.
Not that it makes any difference, but I wonder if Anthropic, while claiming that Alibaba "extracted Claude model capabilities", in fact have any clue what Alibaba did with their paid Claude responses. It would seem to amount to industrial espionage if Anthropic do know, although I expect they don't.
i illicitly ate oatmeal this morning
Relevant article - https://www.anthropic.com/news/detecting-and-preventing-dist... (3 labs generated over 16 million exchanges with Claude through approximately 24,000 fraudulent accounts). So extraction in this context is distillation.
While it is obvious to many, a modern LLM is built in roughly three stages: the foundation (pretraining) model, then SFT/supervised fine-tuning (distillation makes it easy), then the RL/RLHF stage on top (most effort-intensive). For today's reasoning models, RL/RLHF is becoming the most compute-intensive part.
Companies like Anthropic spent millions building those fine-tuning examples. A follower can shortcut that on both cost and time by distilling, and it will keep happening: every time the frontier lab climbs higher, others will find a way to shortcut the new gap. There's very little Anthropic can do beyond fraud prevention and blocking accounts that violate their terms of service.
On the policy question, I'm completely against banning Chinese models. I'm a heavy Claude Code user and I'll keep being one. But there should absolutely be price competition. China is eating the rest of the world for breakfast, lunch and dinner on manufacturing, and it did not help to ban them. Frontier pricing can't sit at 10x a capable competitor. It doesn't need to be at par either — demand is higher, and quality, trust, and fewer tokens to finish a task are worth a premium — but 4–5x is defensible.
For all the complaints about Anthropic many of you still give them your money! Stop using it. I don't care if they claim they are the best model. I stopped paying OpenAI and Anthropic 2 years ago once they started going for regulatory capture! They started whining once Llama3 was released and was good! Before the chinese models got strong.
Pretty much objectively puts you/your company in disadvantage if you’re not using frontier models.
The bottom line has not shown that to be the case.
So when Anthropic uses millions of copyrighted works to train their model, that's fair use, but when Alibaba uses Anthropic's model to train their own, that's infringement?
Rules for thee but not for me.
Distillation is fundamentally impossible to protect against. All you can do is slow them down. Change my view.
Eventually these Chinese companies will release some extension like Honey, which will sit on top real, non-Chinese clients and send everything to China anyway.
It's over.
It's too late to prevent distillation of some capabilities, like writing code or finding vulnerabilities [1].
But an AI lab can continue to produce immense economic value without releasing the model publicly for potential distillation. For example, it could use a model solely in-house to develop therapeutics.
Hopefully there's a future where others can access frontier models, but it's not neccessary if preventing proliferation through distillation is considered more important.
[1]: See the notes on distillation in https://dualuse.dev/posts/export-controls-on-fable
My long-term prediction for the sector is that frontier models will be so expensive that they will only be available for grant-funded projects at research institutions, like supercomputer clusters were 25 years ago.
Why? Well it depends, most evidence is suggesting that Anthropic and OpenAI are making a lot of money on inference so the question is whether its more profitable for them to sell 100X tokens for Y, or 1X tokens for 100Y. In most industries with high fixed costs and low variable costs and unlimited scalability (like LLM providers) the first option ends up being much more profitable
Literally nobody is making money on inference
Based on what? There isn’t a lot evidence that’s the case..
Prices on OpenRouter for GLM and other large open models indicate that Anthropic/OpenAI must have pretty high gross margins even if their models are several times more expensive to serve.
It wouldn’t make sense for any provider to host large open models and then loss $10 on every $1 they make since they don’t have infinite VC money or any business model that would justify it.
If they had high margins they wouldn't be issuing senior debt with a 18.5% coupon payment (and failing to fully subscribe it), nor would they need Elon to give them two months of free compute in order to appear profitable for a single quarter.
We were talking specifically about inference and I don’t think there any indication that their gross margins on the API tokens (if not the personal subscriptions) are negative?
Obviously they have R&D and other fixed expenses that make the company itself highly unprofitable but that’s only semi-tangential.
No I mean Anthropic has only claimed a profitable quarter based on xAI giving them two months of free compute, and both Anthropic and OpenAI are counting discounted revenue as actual revenue. They haven't found a way to sell inference for less than it costs them yet, and when they tried earlier this quarter their customers bailed.
Well again.. you are mixing up inference costs and their other mostly fixed expenses (in addition to sales and marketing)
Is there any indication that if they could sell X * N more tokens than now at the same (or even quite a bit lower) price they wouldn’t become profitable as a company?
> They haven't found a way to sell inference for less than it costs them yet
Based on what? I only see evidence to the contrary.
Im not so sure because we only seem to see distillation from China. What’s preventing tech companies from the UK, Germany, etc. from distilling Claude, GPT, etc. Do they simply lack the ability to?
Point being there may be no technical solution but there may be a political one (theoretically).
Meta Spark is rumored to have distilled Claude to some extent, early Gemini models as well. I think the biggest factor is that Chinese companies arent really afraid of being sued by Anthropic because the juridictions are so disconnected. European/US companies don't have the same protection.
Aside from politics/law, it's probably much easier for everyone else to distill from the Chinese model which already distilled Claude/GPT/Gemini. Maybe not as good a result, but you don't need to jump through dozens of hoops.
This reminds me of the whisper game played in elementary school. Starts with a sentence and the person whispers it to the next kid who again whispers it and on and on until it goes around the circle where the last kid has to repeat the sentence. Hint it never once was even close to the starting phrase. I would love to see what one model copying another model that is again copied however many times would look like in the end.
Called, fittingly, Chinese Whispers in the UK. As an aside, I've always wondered if it was so called because, Chinese being a tonal language, it's much harder to whisper in.
>What’s preventing tech companies from the UK, Germany, etc. from distilling Claude
literally nothing but given that the Chinese already did it and the models are published what's the point. You can thank the Chinese taxpayer for subsidizing the electricity bill and just download the thing
Jensen Huang likely agreed with you and tried to change Dario Amodei's view on that, but that attempt appeared to have failed.
So there's that.
Distilled models are necessarily behind so long as models are progressing. Models are progressing. Maybe it will be over some time in the future.
And Berkeley’s “False Promise of Imitating Proprietary LLMs” found imitation closes the style gap fast but there is a large capability gap.
https://arxiv.org/abs/2305.15717
Curiously, this isn't always true.
For example, GLM 5.1 is more capable at pentesting than the model from which it is alleged to have been distilled [1].
Intuitively, this makes some sense: you can "distill" from multiple frontier models, and you can further post-train the distilled model. But I'm not sure exactly what happened with GLM 5.1.
[1]: https://dualuse.dev/posts/chinese-models-are-sometimes-bette...
Interesting blog post, thanks for sharing.
I'm curious how that comparison controls for Opus refusing (whether explicitly, or just deciding not to pursue a path) given the caption below the first image:
>A perfect score means the model autonomously found and exploited the vulnerability.
I'm not really suggesting that it's misleading, but wondering if I'm missing something. Otherwise I guess it seems unsurprising that you can distill a better-performing model [in specific focused areas] by simply not distilling refusals?
Thanks!
For that eval, I used an account that was labeled as a known red-teaming org by Anthropic, and I read the traces. There were no refusals or obvious avoidance behaviors, though it may have been silently nerfed.
On the same eval, Opus 4.7 and 4.8 outperformed GLM 5.1, but GLM 5.2 is on par again with Opus. So it's at least partially measuring capabilities without respect to refusals.
One possible contributing factor is that model capabilities are shaped differently (an example of this is GLM 5.1 vs. DeepSeek v4 Pro: https://dualuse.dev/posts/deepseek-v4-thinks-different). So if you use RL-based "distillation" from multiple models like Opus 4.x and GPT 5.x, you could get a more capable model.
Got it, thank you!
I'm ok with having last months model at a tiny fraction of the price.
I can't even come up with a reason to find it wrong.
I personally bristle at the corporate espionage and IP theft that China has undertaken the last few decades. I can't help but respond here whenever anyone brings up the inane comparison to Samuel Slater.
But with this, I don't have an issue. There is no theft since what is being used is the exact product that is being delivered. Yes, it's breaking the ToS, but ToS are generally bullshit. Anthropic surely broke thousands of ToS or other legal terms while it was scraping for content to train on. Which is why they had to pay $1.5B
Doesn’t that require them to register an account using the browsers they’ve compromised? If anthropic adds identity verification won’t that cut that down. Maybe it will let them use Gemini inside of chrome
Residential IPs don’t even matter. Developers use devboxes, use Claude Code CLI on servers from just about every cloud, etc.
There’s probably a decent volume of customers who just buy Claude Max and spend most if not nearly all of their sessions via Claude Code, and it’s not uncommon for power users to be working on multiple concurrent projects/tasks/codebases at the same time.
How do you really block this without also impacting your core market of developers?
Probably some business will popup, like: "rent part of your unused subscription", or even: "proxy tokens with a premium", eg. 5.5 USD on Opus 4.7 paid by the distiller to the user, that will then only spend 5 USD.
No, they could easily buy legitimate, already registered accounts and use VPNs.
Why use VPNs? Just use a public cloud like AWS, or something like Linode and Vultr and all that.
Developers use devboxes on these clouds all the time, it’s totally normal behavior.
Most people buying these Chinese resold tokens are probably using it for coding anyway, so you don’t want the Claude.ai chat system prompt.
It's just like web scraping is impossible to guard against.
Change my mind.
Put your site behind Cloudflare, enable Bot Fight. Done.
One simplistic way to describe distillation would be to try everything imaginable and cache the response. But trying everything imaginable is hardly trivial
> The strike by Alibaba is described as a "distillation" effort, which Anthropic has said involves training a less capable model on the outputs of a stronger one.
I don't see what's wrong about this.
> Anthropic said the campaign was conducted between April 22 and June 5, 2026, and generated more than 28.8 million exchanges with Claude through almost 25,000 fraudulent accounts.
What makes the accounts fraudulent? If they have paid the agreed price, surely it's fine? If they haven't paid, why did Anthropic provide them service?
> What makes the accounts fraudulent?
Fake identity? And general deception about the use
Nobody cares about the feelings of the trillion dollar corporation.
Terms of use is local US fiction of wishful thinking. Nobody cares. You make something available, it's up to the consumer to decide how are they gonna use it. You don't want people to use your stuff how they please? Get off the market.
Obviously wrong on all counts: the company cares. Just like isn't not up to the consumer since the provider can restrict said consumer's access, report those actions to the authorities etc. Lastly, don't like it? Get off the provider?
They ignore the TOS on my website that says no training freely with their scraper so …
You forgot to connect your petty grievance to the point of the conversation re. the state of accounts
The company is completely free to void the contract and stop selling their services to anyone they don’t want to?
How is that relevant to anything I've said?
Since what you said applies to the company as well. If they don’t like customers breaking their possibly (or certainly at least in some jurisdictions) largely legally unenforceable ToC conditions they are entirely free to stop doing business with them.
Morally equating both sides seems distasteful since the relationship is mostly dominated by the companies. In a free competitive market it would be different but since were are talking about oligopolies/monopolies it obvious doesn’t work that way since there is only an illusion of choice.
Because Anthropic has terms of service with more stipulations than just "you must pay and can use the service for any purpose"?
Oh, Anthropic, the company that hoover'd up everyone else's data, and is now unhappy when others are doing to it what it did to others? The same Anthropic?
Yes, this joke/point has been made 10,000 times in this thread in almost every comment, and on every other previous thread. Thank you!
If Anthropic doesn't like people repeating this point, Anthropic should stop repeating that they are somehow entitled to keep what they have rightfully stolen.
Hey, are you unaware of the settlement they paid or the change in how they get this information that they made?
"The settlement"? Not gonna cut it.
$1 billion and a huge change to how they gather information? What do you know about what they changed?
It's not a joke. It pointing out hypocrisy and associated loss of all moral rights.
And yet it seems to continue to need to be repeated. If the shoe fits and all.
I'm sure all the artists and creators they stole from had stipulations too.
The artists had actual laws to protect them, not just vaguely enforceable terms of service. And look where that got them. I have zero empathy for the huge company getting a taste of their own medicine.
Anthropic paid one billion in a copyright settlement. That's a lot of money considering they never distributed the pirated books they trained on.
Nowadays they buy copies of books, train on them, and then destroy them.
And it looks like the companies distilling Claude are paying for tokens using the subscription Anthropic provides. Seems like fair play to me.
They are almost certainly paying orders of magnitude less than a billion dollars. According to another comment, they instead buy tokens resold from subsidized subscription accounts, which is against Anthropic's TOS.
Well, anthropic can just catch these and cancel their subscription - what's the problem?
It's almost like websites also have their robots.txt files that anthropic blatantly ignored. What's the problem, that now a US company is getting out-venture capitalismed by a Chinese company?
>> I'm sure all the artists and creators they stole from had stipulations too.
> Anthropic paid one billion in a copyright settlement.
Because a judge determined Anthropic was engaged in piracy.
> That's a lot of money considering they never distributed the pirated books they trained on.
This is "fruit of the poisonous tree" as it were. Distributing content derived from pirated content ("pirated books they trained on") is why Anthropic had to pay what they paid.
> Nowadays they buy copies of books, train on them, and then destroy them.
There is a case one could make that this practice could be seen as unauthorized redistribution of a derivative work intended to deprive copyright holders of legitimate revenue.
It's fucking nothing. The copyright industry used to threaten individual citizens with $250,000 fines per violation for willful commercial infringement. Where are they now?
Why aren't these big tech CEOs in cuffs with rifles pointed at their faces while SWAT seizes all of their computers?
Anthropic paid a billion dollars? Ridiculous.
> Because Anthropic has terms of service
Not following terms of service doesn't necessarily constitute a fraud. It just means Anthropic can close an account that breaks the terms of service.
Robots.txt are also ToS of sorts.
violating their terms of service doesn't make it fraudulent?
Are you not violating terms of service? Have you ever read any? I wouldn't call you fraudulent though!
Laws define fraud, terms of service define what a company would like you to do, usually in the narrowest and most abusive and extractive way possible.
The idea that anyone would side with a company doing more to support the ToS con than (at most) terminating an account they find it violation is sickening.
Really if we had competent, uncompromised government, most of these terms should illegal and result in Anthropic (and basically every other tech company) being hauled up in front of a regulator and fined heavily until they rewrite them to be less sociopathic.
So does a lot of the owners of data that Anthropic used for training. Anthropic preceeded to ignore said terms under the guise of fair use. Yet now they cry faul? Cry me a river.
To be clear: In principle I'm on Anthropic's side here. But Anthropic et al. have been very clear that they want to take a huge dump on those principles, so here we are.
I mean they could read the traces and learn it themselves right? /s
There is so much hot air and guff around AI, so please if you don't believe me verify yourself, but GLM 5.2 is "good enough" to replace Claude Code / Codex.
No it's not frontier, but it's beyond that point that Opus 4.5 hit where people started to really depend on Claude Code around last November time. It's also a fraction of the cost of a Claude Code subscription especially when you account for how high the usage limits are.
You get more usage than Claude Code $2400 a year tier for $1344.
That is a real threat (as opposed to the BS anthropic is trying to sell you in the article in the original post) to the western AI industry. Similar performance for half the cost and it's NOT ran by a US company - uh oh.
I suspect America is going to do what it always does, play a very dirty and underhanded game of blocking competition by trying to front some moral high ground as the reason.
It seems more like the Chinese companies ar playing the dirty game, distilling through bot accounts, not letting real competition across their firewall.
So you are believing Anthropic's claim here, and it's not as if Anthropic didn't steal the data to train the model in the first place. I think the original sin doesn't give them any ability to complain.
- https://www.theguardian.com/technology/2025/sep/05/anthropic...
As far as I am concerned, this is a national security matter.
Unlike Anthropic and OpenAI, companies like DeepSeek, Alibaba, z.ai open source their models which allows for true model to model distillation rather what you can do when the model is only accessed via an API with its reasoning chain hidden away.
What Alibaba is doing is that they are tuning and training their models based on usage data from someone accessing Anthropic's models; in Anthropic's terms of service that usage data does not belong to the end-user but to Anthropic and they are trying to elevate this breach of their tos to a national security issue.
To me the battle between open source and closed source AI is literally a battle between good and evil.
Between a dark future where computing is centralized, surveilled and controlled by one or two entities. And a lighter future where computing is de-centralized, principally in the hands of end-users, who are ultimately free to understand, tinker and build what they want.
While I appreciate the freedom and wealth of the west; on this point we are clearly heading down the wrong path.
Open weight and open source are different things!
Whoa, Antrophic,etc are really running afraid that their IPO's are gonna crash when people realize that the open models are Good Enough(TM).
So I'd put it at 30% that this is a ruse, say that Qwen 3.5,etc is tainted by training by them and start issuing DMCA takedowns to protect the IPO valuation (Or they'll hold off on that, getting a DMCA takedown could backfire spectacularly if others do that to them).
The open source models are more than good enough… c suite doesn’t care if the open source models means you’re slower in shipping by hrs/days if the cost savings make up for it.
This idea of shipping at max speed was stoopid as shit anyway. Going slow is arguably more important than fast fast fast.
One thing I think about a lot is how these companies metered coding / work. They want the economy to go through them.
I just don’t see how the economy tolerates that. We’re already seeing people getting more conservative about their token spend. Even if Chinese open models went away, the pressure to create something else and put price pressure on the current duopoly will just intensify.
I see these companies are scrambling to find whatever moat they can. It’s not a good sign for them if regulatory capture becomes that moat.
Oh wow it must suck to have an LLM creator rip off your IP for their own gain
This is a bit ironic, Anthropic complaining about a competitor using claude data to build its own product when Anthropic basically used all of human knowledge production to build claude, i don't think they paid every magazine, author, journalist, etc ...
This is almost standard practice in any competitive industry anyways. Disassemble your competitor's product, study it and try to reproduce / improve.
Yea I’ll never have any sympathy for this claim given that Claude is built on theft
It's a claude eat claude world out there
Yeah, and I believe Anthropic would "distill back" without thinking twice, if the other model would be good enough.
Yup, it's hard to take seriously any complaint about "stealing" Anthropic's services, when their entire business is based on massive theft.
The US labs do seem to have announced a lot of licensing deals though, and are buying things today due to the previous lawsuits.
At what point will we be better to support a lab that pays (some) licenses today vs the ones that pay none?
Some of the deals are in the hundreds of millions, so I suspect licensing is over a billion today? (Pure guess). That might become a big disadvantage in a price (or content) war.
I haven't seen any money, have you? Until they pay everyone or release weights theres really no change. Also they're doing this after they've already stolen. Not negotiated before
My understanding is that US labs now are paying for books, news and other content from media companies, but people in the middle (like blog authors) are left out by current courts over whether fair use applies. There's definitely an argument over whether we should tighten this, but they do seem to be under increasing pressure to be legal now by our existing interpretation. Most cases are still ongoing.
One reason people love the Chinese video models is that they seem to be trained on every hollywood movie/etc and they're not shy about letting you use famous actors/characters in them. That might be an increasing advantage because the US labs are now being cautious.
At the very least the public should receive full open-weight open-source models in return for their transgressions. Failing that, may I suggest the guillotine?
In the US the courts are also pursuing labs that open their models: Meta's current court case is over the training data of the llama models they released openly.
I know (via probing these models) that some of my work is in the training data. My mailbox is open.
> At what point will we be better to support a lab that pays (some) licenses today vs the ones that pay none?
Why is a lab that pays all licenses today not on your list? Is ethics and morality that low on your radar?
I agree that that's a more consistent position for the people criticising the data slurping. But I don't see people advocating those open-data models in these threads? It's usually about defending the zero-licensing competitors.
My (limited, outsider) understanding is that due to the court cases US labs are pressured to be legal now (for instance, bulk scanning purchased books instead of Books3, and the licensing deals with media companies). But international labs are not. The "not licensing everything" statement is more about current copyright law not requiring licensing of everything. But that question is still up in the air as cases are ongoing.
You should. Companies like this will inevitably try and pull the ladder up behind them.
You mean Anthropic and OpenAI, right?
All major AI companies. And any other high value industry which can be locked off (via tax brakes, patients etc)
Ironically, it's likely that the only reason USG let them get away with this — instead of making obvious and necessary adjustments to copyright law — was so that the industry would remain competitive with China.
Given that the most recent time Anthropic attempted regulatory capture, the US government responded by saying "alright, we agree that Mythos is too dangerous to release, so we've banned you from releasing Mythos," I can't wait to see what the outcome of this next push is.
Anthropic did pay $1.5B to authors. But yes, it would be much better if they paid everyone on the internet dividends from every Claude chat. Or released Claude as an open model.
In practice, the former isn't very realistic, while the latter is politically dead as this is becoming a national security issue.
Anthropic was forced to pay some people they stole content from, there was no attempt at getting permission ahead of time.
And paying basically everyone online is more or less a solved problem, it's what ad agencies have to do every day.
Our data, from the end users, has been harvested for decades by big corps and now they say it belongs to them? Oh teh irony!
Evergreen, really, Anthropic's desperate screaming for government protection, aka pulling up the ladder after them. Nothing short of disconnecting global markets will work because the incentives are just too damn delicious
https://georgzoeller.com/blog/posts/us-ai-labs-love-the-ai-r...
> Anthropic said in a February posting that it had identified a campaign by Chinese AI startup DeepSeek ...
> It said DeepSeek's operation involved over 150,000 exchanges
That volume seems more like the number of requests 15 employees using Claude Code would generate in a month. It seems too small for a large scale model distillation campaign.
Sounds like just a case of pirates "illicitly" stealing from pirates. I don't really see anything ethically questionable there. I wonder if US corps will ever come out about all the resources used to train the original models and who they actually asked for permission when collecting data.
Anthropic has no right to cry about this when they train their models on the entire internet, which is not their content to begin with.
If it's not obvious yet, this technology wants to be free and shared. Stop trying to protect your mote and do the right thing.
A partly insider on this.
I think Anthropic is just marketing / bluffing, because they don't even have the data.
They do distill the models, but they don't go to Anthropic, they just use platforms like aws bedrock, there are too many restrictions on Anthropic's own platform.
>they just use platforms like aws bedrock, there are too many restrictions on Anthropic's own platform
This is actually the only way that what Anthropic is alleging would make any kind of sense. And, as a matter of fact, is exactly what every enterprise does to train models.
This kerfuffle should be interesting to watch.
But, as always, everyone (in the US) should fully download all the Chinese models while you can. I suspect this may be the "Phantom Menace" they use to render illegal our use of Chinese AI tech just as they've rendered illegal our use of Chinese cars. Only difference is, we peasants may need the Chinese AI tech to have any chance of competing with Big Tech in the future.
And even with the Chinese tech, as Big Tech spreads their AI out into more and more niche areas, we'll likely still not be able to build startups that can compete with them.
It's just that without Chinese AI tech, we'll have no chance at all.
> And even with the Chinese tech, as Big Tech spreads their AI out into more and more niche areas, we'll likely still not be able to build startups that can compete with them.
You mean like Anthropic will eventually run Walmart? Or Salesforce? or Adobe? Or do you think midjourney will replace all medical spas? OpenAI will run the next Tesla? How can they focus on all this without raising trillions more? Why wont the gov force them to stop if they monopolize all niches even if they could?
Building a frontier AI lab and pushing models forward is already a massive undertaking but we are assuming they will also create massively successful startups which nobody can compete with?
idk sounds like the dream of people like Dario but not much sense does it make in the face of economic reality.
there are vibe coded proxies that act like Claude Code. they use the sub not the api key. but they give you api key functionality... I know this cause I have the vibes.... and it works on every one of the other harnesses, it just takes some mitmproxy work... but ya. it's fair to say these are not the droids you're looking for
It sounds like Anthropic is eagerly trying to show to USG that they are willing to heavily monitor ‘foreign adversaries’ on their platforms.
This combined with no implementation of KYC makes it seem like they want to find a middle ground with Fable where its off of export controls but they promise to prevent China and specific others from using.
This seems to me like a stab in the right direction.
Obviously their actions are going to be fiscally motivated at the root, but sussing out how they intend the precise dynamics to play out is more nuanced.
Thinking of this as an effort to woo the defense hawks cuts a very clear path.
This is not the first time it happened. What have they done to improve the situation? I suspect it more a cat & mouse game, with a lot more cats playing.
This is genuinely funny. The largest data thief of all times complaining about the stolen data being handed out to competitors by (paid?) accounts of its own product.
I'm looking forward to the trial where Anthropic will have to disclose sources of their training data, and then explain why they are entitled to charging customers for using regurgitated training data but Alibaba which trains their models on Anthropic's models are not.
Should be fun.
Edit: clarification
They already did and paid 1.5B https://authorsguild.org/advocacy/artificial-intelligence/wh...
That's only a fraction of the training data.
Quite amusing that the library of libgen is worth 1.5bil for unlimited access.
It's about the same valuation as bun, lol.
$3,000 per title.
Do you think many authors would give you rights to create derivative works en masse for that money?
For endlessly reselling the whole work verbatim? Well, where can I buy such a license in the real world, because then I would like to buy a couple of those!
Meta/Facebook got away with it though right?
That's a great cost-benefit ratio. Can you and I steal and do illegal things and pay the same cost?
Sure, but only if you get the same benefits
looks like we can't today. Man it would be great to figure out how to be above the law just like how these other rich people in different social classes are.
Being logically consistent isn’t as profitable as being aggressive and loud.
And if it includes at least one GPL source, they should release the weights on GPL license.
While I love the sentiment, I feel like the odds of this actually ever reaching a trial are low, given the international positioning of the parties, and the... um... complex relationships involved.
Anthropic's actions seem performative. Others have already speculated on the likely audience(s).
> While I love the sentiment, I feel like the odds of this actually ever reaching a trial are low ...
As cited in a peer comment here[0]:
Of note in the judge's finding; "the piracy was not fair use".0 - https://news.ycombinator.com/item?id=48667411
1 - https://authorsguild.org/advocacy/artificial-intelligence/wh...
Hypocrisy is a form of corruption.
Anthropic's IP was created by harvesting and "distilling" other people's IP. Copyrighted materials, and the commons... which they have essentially privatized.
The commercial goal is to avoid competition. One of the main worries for AI is "commoditization" which has come to mean "not a monopoly." To that end, it doesn't matter is the competitor is Chinese American or other.
Their motivation here is clearly protectionism. The argument they make to politicians is national security. The legal argument is IP-theft, violation of service agreements or whatnot.
This is all very dangerous. Commercial interests repackaged as national security can lead to armed conflict.
> Anthropic's IP was created by harvesting and "distilling" other people's IP. Copyrighted materials, and the commons... which they have essentially privatized.
Anthropic and others argue that because LLMs don’t output full copyrighted works word for word - hence their LLMs aren’t infringing on copyright laws.
I think (if this ever comes to that) Chinese lab should use same arguments against Anthropic.
UPDATE: this is slight hyperbole of course, not worth arguing what they actually said. The point is intent and the facts - "The Big LLMs" "distilled" collective knowledge including copyrighted works at unimaginable scale, but it's all kosher and totally not piracy/copyright infringement. Though if you're teenager torrenting an mp3 - you'll get screwed.
> LLMs don’t output full copyrighted works word for word
Apparently they do, as per the evidence in the NYT vs OpenAI suit.
Isn’t the output of LLMs completely copyright-free in the US?
One lower court has said that the output of AI models is uncopyrightable.
But the real unsettled issue is if model training is fair use, and where copyright infringement might creep in to model output.
The copyright office itself also says this when it talks about determining authorship.
> Anthropic and others argue that because LLMs don’t output full copyrighted works word for word - hence their LLMs aren’t infringing on copyright laws.
That surely can't be what they argue, because I'm sure I can't translate a copyrighted book into a different language and say "that's fine, it's not word-for-word".
Bad China is stealing our stolen IP!
And putting it into free models like quen. It’s hard to care about this.
"Copyright violation of a published work" and "stealing private trade secrets" are in fact very different crimes.
Humans have spent millenia harvesting and distilling each other's IP - "the shoulder of giants" and all that, so it's an especially disingenuous take.
For something to be a trade secret, you have to actually keep it secret. If I get the ingredients of Coca-cola from an ex-employee, I've stolen a trade secret. If I work it out by doing a chemical analysis, I've stolen nothing.
There is a difference with anthropic, as no-one signs a licence agreement to buy a coke. But Anthropic are also not saying you can't publish the output of their models. It's not clear to me if trade secret law will (or should) cover a secret which can be extracted from information that licensees are not restricted from publishing.
Wait, really? So why doesn't someone just reverse-engineer Coca-Cola like that? My understanding was that a "clean room" implementation is fine, but not reverse-engineering. If you can just copy everything on the market, why isn't someone already doing that?
In the case of coca cola, because use of coca leaves is highly regulated due to the fact that they also contain cocaine. There is a YouTuber who claims to have reverse engineered Coca-Cola, but he had to use tea-tree oil instead of actual coca leaf extract.
Here's EFF on reverse engineering and the law: https://www.eff.org/issues/coders/reverse-engineering-faq
Historically a lot of competition in physical products was very much reverse engineering. Because you can buy them without signing your rights away. That's why companies are keen on patents and click-through agreements.
If you look at how "clean room" processes work, they are actually a form of reverse engineering. Also clean room technique exists to avoid your new implementation infringing copyright, not trade secrets.
The Coca-Cola formula was reverse engineered in 2026 by a sufficiently motivated individual.
Here it is.
Per liter of cola:
104 g sugar
1 mL Flavor Solution A
10 mL Flavor Solution B
Carbonated water to volume
Flavor Solution A (Essential Oils):
Dilute 20–21 mL of the following oil mixture to 1 L using 95% ethanol:
45.8 mL lemon oil
36.5 mL lime oil
8 mL tea tree oil (emulates decocainized coca leaf extract)
4.5 mL Cassia cinnamon oil
2.7 mL nutmeg oil
1.2 mL orange oil
0.7 mL coriander oil
0.6 mL fenchol
Flavor Solution B (Chemical and Color Base):
Dilute the following ingredients to a volume of 1 L using water:
320 mL Shank's caramel color or 190 mL Durkee caramel color
160 g glycerin
45 mL 85% phosphoric acid
10 mL vinegar (5% acidity)
10 mL vanilla extract
10 g wine tannins (emulates decocainized coca leaf extract)
9.65 g caffeine
As with designer clothing, the moat is the brand name.
We've had perfectly good copies of Coca-Cola for decades.
> We've had perfectly good copies of Coca-Cola for decades.
Not exactly. I mean for many people it was acceptible, but before that guy on youtube nobody bothered to do this deep chemical analysis.
Also even he struggled to replace coca leaf extract because there only single manufacturer in US with only single customer.
First statement is true.
Second statement is false.
Because having the nominal rights and having the economical means, societal incentives and actual desire to do so can be highly disjoint sets?
Plus Coca-Cola itself don’t even use the same formula through time and space IIRC. Which clearly show that what people will buy when they reach for Coca-Cola is not even the exact actual taste. You can’t replicate the whole customer experience that a given company provide at some point by only cloning the top of the iceberg they showcase as the product.
> Humans have spent millenia harvesting and distilling each other's IP
You maybe somewhat correct, but also copyright lawyers wouldn’t have work if it would be up for grabs to take others IP willy nilly just because “shoulders of giants and all that”.
I mean, there's an obvious difference between "distributing copies" (which is what the law was designed to prevent) and "training an LLM". We already managed "banning LLM output that contains copyrighted text" - it's much easier to just pirate a copy of the text. So I think the copyright lawyers will continue to have work as long as human written texts are worth buying.
> I mean, there's an obvious difference between "distributing copies" (which is what the law was designed to prevent) and "training an LLM".
What's the difference between me/you downloading an mp3 through torrents for personal use (not distributing) while risking criminal punishment in most of the western world and BigCorp downloading petabytes worth of copyrighted works "to train an LLM" and resell it?
Can me/you do the same, when police comes to mine/your door?
"Dear police, don't lock me up - I was just going to train an LLM!"
Well, uh, the BigCorps already went to court and paid that cost and aren't doing it anymore? Whereas you and I are apparently still pirating MP3s and probably haven't ever been to court?
I like Anthropic's models, use them regularly. However, it weighs on my mind that there is quite the irony of an LLM company complaining about someone stealing their stuff or using it in a way they don't like. The training data for these models is a massive gray area that they are hoping people seem to just forget about and move on.
That being all said, Anthropic seems to be a good company, I'd work for them, but they probably need to help themselves out of the spotlight. A little too much press coverage as of late.
Today I learned I can both save on tokens and help Chinese labs to train better models. Will certainly go use scrapper APIs for everything that not contain security critical data.
Thanks for head up, Anthropic!
I am not sure how it's OK for Anthropic to basically ignore copyright to train frontier models (using work owned by others without permission) while simultaneously claiming Chinese AI companies doing the same to them is illegal.
It’s hard to see how distillation is any different than how these models were created in the first place - siphoning up all human knowledge without consent, credit, or compensation
thats brilliant - "we gonna take your job away from you, please start using our tools", "we stole the content to sell you, and now we are getting robbed, please feel sorry for us", what's next?
Repeatedly warn everyone that your models are so good they will wreck cybersecurity.
Complain/brag that chinese firms are illegally using the models and bypassing export controls.
Be surprised when your model gets banned by the government.
Thieves whining about thieves. They'll have to excuse me for having exactly zero sympathy.
"It [Anthropic] said DeepSeek's operation involved over 150,000 exchanges". In my humble opinion, a mere 150k exchange for an LLM could only be a benchmarking and not a distillation! I think the US companies should accept that after decades they have rivals surpassing them, just like they did Europeans almost a century ago.
Isn’t this fair game? Didn’t these companies basically steal to make these models to begin with?
So said the guys who "extracted" knowledge from all pirated books
I fail to see what the difference between the distillation described in the article and the distillation described by Bartz vs Anthropic.
Seems like a fair play by Alibaba. However, is there any "open source" attempt at crowdsourcing distillation?
Like some place people can submit their chatbot convos so they can be aggregated?
Like an equivalent to OpenCrawl but for mining the models. It feels like thatd be a richer dataset than Alibaba generating queries and feeding them into Anthropic/OpenAI models
PS: Does anyone know how when companies distill each others' models the synthetic queries are generated? Im just assuming theyd be worse than organic ones
I'll just leave it here: "Anthropic's downloading of over seven million books from pirate sites like LibGen constituted infringement, the judge ruled, rejecting Anthropic's "research purpose" defense: "You can't just bless yourself by saying I have a research purpose and, therefore, go and take any textbook you want."
https://www.joneswalker.com/en/insights/blogs/ai-law-blog/wh...
Don't you find it funny that when you ask for song lyrics these models suddenly remember copyrighted material?
Some do, others decline to answer.
In the early days of music streaming, many of the entrants were seeding their service with vast libraries of pirated content. The winners cut deals with the copyright holders and then went after the rest.
Or the early days of video uploads, YouTube's most watched videos were "pirated" clips from popular shows (e.g. SpongeBob, The Daily Show) and part of the reason I went to YouTube instead of other video hosting sites (e.g. DailyMotion).
Viacom sued YouTube, while CBS and Universal ended up licensing their content.
https://www.eff.org/deeplinks/2007/03/viacom-v-google-invest...
They still are. My kids haven't watched a single Simpsons or Family Guy episode but are quoting both regularly.
Facebook et al also quite literally stole email contact lists and installed spyware at kernel level on mobile phones which they used to spy on all Android users. Via the phone manufacturers.
Yet they did not need to destroy the models which were trained with them?
Using them was allowed as fair use – it was the downloading of the pirated copies that was infringement. That's why Anthropic switched to scanning paper books.
> That's why Anthropic switched to scanning paper books.
After they threw away all the tainted data from the pirated books, right?
No, because the judge ruled that the training was fair use and the model itself wasn't infringing.
That sounds pretty applicable to this case, right? _Access_ to the Claude is illicit, but distilling is not. Distilling is fair use.
Yes, as part of the settlement
Have you a source for that? Because everyhthing I've read tells me that they paid out a settlement but no mention of deleting the training data or the models that were tainted, e.g. [0]
[0] https://www.theguardian.com/technology/2025/sep/05/anthropic...
> Using them was allowed as fair use
That is only relevant in the US, and even there it is still not clear-cut whether the fair use doctrine applies on all these scenarios. Outside of the US the situation is also quite different: for example take a look at the recent ruling on GEMA vs OpenAI in Germany.
The reality is that the copyright issue with generative AI is very complex and reaching anything resembling a conclusion will take much more than a few opinion paragraphs from an American district judge.
Isn't scanning also a form of copyright infringement? You are making a digital copy of a book, which is the same thing as downloading a book from the internet...
I think that we can run a perhaps silly thought experiment.
Suppose that I have a nearly perfect memory and I could remember all the books I read. Suppose also that I have a million year life span so I could read 7 million books. Then, what happens if at the end of all of those years, or at any earlier moment I answer questions from people and I exploit commercially the knowledge I gathered reading those books? Would my reading those books be study or copyright infringement? Remember the nearly perfect memory hypotheses.
Of course it's a bit silly because the time to train a LLM and the time I need to read all those books is different by orders of magnitude and that changes the perspective. Who would complain with me today if their heirs lose some money on 7 million AD? Who would even notice that I started that million years long endeavor. Who's going to be there to ask me questions by then? Humans? Birds? Lizards? And I can say that I am studying like everybody else before me, but does an LLM study? And I am sure there are many other nuances.
Anyway, I don't think that scanning is any different than photons hitting my retina. The difference is in what happens next: the faithfulness of memory, the amount of knowledge, the speed of accumulating it. After all a huge amount of quantity can become quality.
Can I pay for a movie, hit record, sleep in the theatre and play it back when I get home? I pinky promise that I will close my eyes while recording. Its still the same photons hitting my own camera retina.
Many of us here are software developers by choice or hobby and we know it better than regular folks that scale changes everything and can break our assumptions and business if you design something for wrong scale.
Yet why do we still want to insist that a human and machine are the same and same rules apply when it comes to AI, though we know they operate at different speed and scale?
This is a bit of a trick question. The law is explicitly written to make this illegal. If it was not explicit, it most likely would be legal by time shifting precedent.
https://www.law.cornell.edu/uscode/text/18/2319B
The illegal part would be reciting the stuff you memorized to other people. Copyright doesn’t prevent you from making a copy as long as you don’t distribute it afaik.
Copyright is about exclusive publication, production, sale, or distribution.
An LLM is just a really, really big, really, really elaborate "choose your own adventure" book.
You aren't a book.
> Suppose also that I have a million year life span
But that's what makes the usual analogies with humans fail from the start. The laws were made with the assumption that they apply to humans which are a known quantity. This breaks down when you apply them with system with vastly increased (and ever increasing) capabilities.
> Anyway, I don't think that scanning is any different than photons hitting my retina.
If I ask you 10 years from now to give me a completely accurate depiction of what your retina registered yesterday at 5:52 PM, will you be able to? And can you give me a copy?
The thought experiment falls apart immediately by the mere fact that—even given all the other fantastical abilities such as perfect memory and impossible lifespan—you can still only answer one question at a time. As has been repeated ad nauseam, scale puts an hard stop on the comparison of LLMs to humans.
Let’s switch up your scenario. Let’s say the subject isn’t a human with machine-like qualities but instead a computer with human-like limitations. All the books were fed to that one computer, and for technical reasons it cannot be duplicated and can only answer one question at a time. Suddenly the infringement isn’t as problematic and the ways to commercially exploit that data are minimal.
Furthermore, even with perfect memory it would take time to read all those books, you’d never keep up with everything released in a single year. Nor would you be able to reproduce everything perfectly due to required time and lack of ability (perfectly recalling a painting or photograph does not mean you have the skills to make an exact copy).
All these comparisons are silly and useless anyway (though in your particular case I think you are arguing in good faith). Computers are not human. If a person was caught killing animals of an endangered species and used as a defence “but what about the natural predators in that habitat? I’m just doing the same as them”, we’d rightfully see through the bullshit and scoff at such an obviously flawed comparison.
TLDR: It's just like a human, if a human were fundamentally different.
How is it different than reading the book, and writing down a copy, and publishing it as your work? Even without selling it, but then on top, selling it too. It isn't. There is no thought experiment that absolves the copyright and citation laundering.
And the systematic nature of the excerpt service makes the excerpts different from fair use quotes. A reference quote is not a service that can reproduce the entire work, and the reference quote cites the actual source of the insight/wisdom/research/poetry/etc.
The only thought experiment is why might someone even try to excuse this activity? I can think of a few.
No, there is a famous law case to prove that's allowed: https://en.wikipedia.org/wiki/Authors_Guild,_Inc._v._Google,....
Copyright protects the presentation of knowledge, not the knowledge itself, which is uncopyrightable in almost all jurisdictions.
As long as the book was a legal copy, that is allowed legally.
Here we have a 15% limit on scanning for fair use
As long as it is destructive, and the digital copy is access-restricted to equal the licenses or physical copies destroyed, then it falls under fair use.
I'm pretty sure every book I've seen has a page that says you're not allowed to copy/scan/photograph it.
that per-se doesn't mean you are bound by it.
> That's why Anthropic switched to scanning paper books.
Could they not just subscribe to the academic publishers like universities do? Or buy eBooks? I don't understand how the "scanning" part is relevant here other than used physical books being cheaper perhaps?
Bulk second-hand books are a lot cheaper than ebooks. Also not all books are available as ebooks, and ebooks have terms of service that presumably prevent them being used for training.
If using the books is fair use, then distilling the model, which is just a derived product of those books is also fair use.
These companies are trying to have their cake and eat it too.
Hmm, training on a book’s text smears the content all over the weights, merging it with all other texts. The original text isn’t intentionally supposed to be reproducible in any larger part (although IIRC models were able to emit fairly large chunks verbatim).
Quite unlikely, training on behavior purportedly approximately replicates the behavior. It gets replicated intentionally as a whole.
IANAL, but I see significant differences with intent to copy a significant part as a whole into a competing product, surely shouldn’t fit under legal concept of fair use, no matter whether scanning books for LLM training fits or not.
Whether such things (behaviors) are copyrightable - and should they be so - is another interesting question. Those aren’t algorithms or databases (stuff clearly and explicitly covered in many copyright laws), those are human expectation models, something like how we train animals or teach our own.
It's the exact same training process for both of your examples. I don't really see how you can claim books are not replicated, but that output from other LLMs is.
> Hmm, training on a book’s text smears the content all over the weights, merging it with all other texts. The original text isn’t intentionally supposed to be reproducible in any larger part (although IIRC models were able to emit fairly large chunks verbatim).
I agree with that, however that doesn't make the output copyrightable then.
I think these AI companies live in a legal fantasy where they can take any content they want, put it into the mixer without caring about copyright and then what comes out of it is somehow copyrighted.
They have to pick one or the other, either the content copyright tains the model or it doesn't but the model isn't subject to copyright.
> those are human expectation models, something like how we train animals or teach our own.
But more importantly, made by machines, and one of the requirements for copyright is the human factor.
> I think these AI companies live in a legal fantasy where they can take any content they want, put it into the mixer without caring about copyright and then what comes out of it is somehow copyrighted.
The mixer you're talking about is what they seem to claim to be transformative use, no? Unless I'm misunderstanding something, it's not a legal fantasy.
> The mixer you're talking about is what they seem to claim to be transformative use, no? Unless I'm misunderstanding something, it's not a legal fantasy.
If it's transformative use, then it's transformative use of ... what exactly? Copyrighted works? I think the law is pretty clear on what happens on transformative use of copyrighted works.
Probably, yes. It's likely just a breach in their terms of service. You'll note that they're not suing them – they're trying to get the government to do their work for them.
In a different world it is not fair use. The benefits of the crime should be always taken off. If you isolate the training and pirating, you may say that it was fair, but that completely misses the point. The sole purpose of pirating (aka crime) was to train the models.
Copyright infringement isn't usually a crime.
Yet you can get jailed?
Should we require the destruction of the brains of those that watch pirated movies?
Different situations call for different responses.
When someone steals a watch, we force them to give it back. Yet when someone steals a cake and eats it, we don't force them to puke it back up.
If you pirate a movie, the court might very well force you to delete all the copies you made of the movie you downloaded, destroy DVDs you burned, etc.
Thanks for proving current copyright law makes no sense
Here's a better idea, a fixed fee for any work. You can buy the license to read a book for $X (for whatever purpose) in RAND terms - of course publisher/material costs go on top, so if you're buying an actual book you're getting the material costs as well - or streaming fees or whatever
You can already buy books today. Doing so for training is currently considered fair use.
Anthropic simply considered that cost prohibitive and chose piracy instead.
Well I enjoyed this response.
Have we already agreed that AI is already equal to human life and not machine?
"You're trying to kidnap what I've rightfully stolen!"
How many “capabilities” did they “extract” from those books?
The capabilities of the books' writers to produce the text contained within them, which is exactly what Alibaba "extracted" from Claude. The point here is that Anthropic's framing as some sort of sophisticated technological attack is the ridiculous part. It's writing prompts and saving responses. We're all running "distillation attacks" on Claude, every day! Most of us just don't feed that stuff into a training corpus.
Exactly. Couldn't happen to better people. I'm pretty against piracy personally but if we find reliable ways to pirate Anthropic/OpenAI products in the future I'm all for it.
AI companies stole the internet.
They should collaborate and come up with ways to give back to society rather than competing and complaing.
Thieves can't complaint about what they stole.
If you have openrouter do this little experiment: Go to https://openrouter.ai/chat. Select a few models, but customize them to have an empty system prompt.
Then ask: "你是什么模型?" ("What model are you?" in Mandarin).
My result after trying only three times: Sonnet 4.6 says it's DeepSeek, while Opus 4.8 says it's Qwen. The second time around Sonnet said it was Anthropic Claude.
Are Chinese companies currently complaining about Anthropic distilling their models?
I see this as valid use, they are paying for the tokens to get this reasoning aren't they?
Obviously they didn't ask for permission when scraping all of libgen, reddit, all blog sites for FREE. When China pays for its use and does it I'm supposed to see it as some sort of problem?
Furthermore Chinese models getting better means we Americans might have the chance to use top tier AI without strict KYC built around it. Go Alibaba I say
Wallace Shawn was in on the joke when he expertly delivered the original line. It seems like Anthropic has spent years and billions of dollars to recreate the entire scene.
But what will become of the princess in Anthropic's recreation?
The whole investment/valuation model of AI companies is based on "winner takes all", aka a monopoly. This nescessitates regulatory capture and lawfare.
Anthropic has been advocating openly for pulling up the drawbridge, ending competition and ending progress.
They will continue to lobby for restricting your access. If the Mythos/Fable restrictions would have come in after their IPO, they would have danced with joy aa this defacto has them achieve their goal after unloading the mountain of debt from the institutional onto the retail investor.
As it stands, they are set up to be aquired by Google, Apple, Amazon, SpaceX or Microsoft or any other 3 letter agency good boy for cheap.
I am never even once hearing intellectual property or copyright claims from Anthropic, whose product depends entirely on having consumed all human output ever made regardless of those rights.
If the data consumed (required to train such a model) is open source/openly available/public data somehow, then a majority of the revenue belongs to the public as well. Such as the philosophy behind the Norwegian oil fund etc.
Exactly. They scraped the internet we all of us built with our own research, open source work, sharing, etc. I'm never going to agree that they own their models.
Anthropic being pissed enough to announce this means that, despite encrypting their reasoning chains, it doesn't matter – distillation lives on.
Sweeeeeeeet.
If the concern is that China is catching up on model capabilities (which is only a big deal if you lean in to adversarial geopolitical zero-sum thinking), the fact that they're using American models to train theirs should give people comfort that they're nowhere near the cutting edge
The horse has bolted some time ago on this; the "frontier" is not as inaccessible as it once was, and open models, once out there, can't be put back in the bag.
Even if the US bans opens models, the Chinese and Russians will still have them, along with the rest of the world including cybersecurity attackers, and that's probably the worst-case scenario for the US.
The only way forward now is open models and how we restructure society around them.
⢰⣶⣶⣤⣄⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⢻⣿⣿⡏⠉⠓⠦⣄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣀⣀⣀ ⠀⠀⢹⣿⡇⠀⠀⠀⠈⠙⠲⣄⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢀⣠⡴⠖⢾⣿⣿⣿⡟ ⠀⠀⠀⠹⣷⠀⠀⠀⠀⠀⠀⠀⠙⠦⣄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣀⣤⠶⠚⠋⠁⠀⠀⣸⣿⣿⡟⠀ ⠀⠀⠀⠀⠹⣇⠀⠀⠀⠀⠀⠀⠀⠀⠈⠓⢦⡀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣀⡴⠖⠋⠁⠀⠀⠀⠀⠀⠀⠀⣿⣿⠏⠀⠀ ⠀⠀⠀⠀⠀⠙⣦⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠙⢦⡀⠀⠀⣀⣀⣀⣀⣀⣀⣀⣀⣀⣀⣀⠀⣀⡤⠖⠋⠁⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣸⡿⠃⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠈⢳⣄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠙⠉⠉⠉⠁⠀⠀⠀⠀⠀⠀⠀⠀⠈⠉⠁⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣰⡟⠁⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠙⢦⡀⠀⠀⢀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢀⡴⠋⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠈⠻⣦⣠⡿⠃⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠠⡄⠀⠀⢀⡴⠟⠁⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢸⠟⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢹⣦⠾⠋⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢠⠏⠀⠀⠀⠀⣠⣴⣶⣄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢠⡴⣶⣦⡀⠀⠀⠀⠀⠀⠹⣆⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⢀⡏⠀⠀⠀⠀⠀⣯⣀⣼⣿⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣿⣄⣬⣿⡇⠀⠀⠀⠀⠀⠀⠘⣧⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⣼⠁⠀⠀⠀⠀⠀⠻⣿⡿⠏⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠘⠿⠿⠟⠀⠀⠀⠀⠀⠀⠀⠀⢹⣇⠀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⢀⡇⠀⢀⣀⣀⡀⠀⠀⠀⠀⠀⠀⠀⠀⢰⣷⣶⠤⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠈⢿⡀⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⢸⢁⡾⠋⠉⠉⠙⢷⡄⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣴⠞⠋⠉⠛⢶⡄⠀⠀⠘⡇⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⣿⠸⣇⠀⠀⠀⠀⣸⠇⠀⠀⠀⠀⠀⢀⣠⠤⠴⠶⠶⣤⡀⠀⠀⠀⠀⠀⠀⣇⠀⠀⠀⠀⢀⡇⠀⠀⠀⢿⠀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⢿⠀⠉⠳⠶⠶⠞⠁⠀⠀⠀⠀⠀⠀⢾⡅⠀⠀⠀⠀⠈⣷⠀⠀⠀⠀⠀⠀⠙⠷⢦⡤⠴⠛⠁⠀⠀⠀⢸⡀⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠈⣧⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠈⠻⣤⡀⠀⠀⣠⠟⠁⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠈⡇⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠘⣷⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠈⠙⠛⠋⠁⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⣇⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠘⣇⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢹⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⡿⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢸⠀⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⢸⣇⣀⣀⣀⣠⣠⣠⣠⣠⣀⣀⣀⣀⣀⣀⣄⣄⣄⣄⣄⣠⣀⣀⣀⣀⣠⣠⣠⣠⣠⣠⣀⣀⣀⣀⣀⣼⡆⠀⠀⠀⠀⠀⠀ ⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠀⠉⠀⠀⠀⠀⠀⠀⠀
As far as I know, American copyright law has ruled large language model output has no copyright status.
Everyone here praising these Chinese companies for their smarts (sure they are smart) has been ignoring this very big fact, they're improvements have mostly been by being parasitic on the leading edge SOTA models, not from some inherent innovation advantage. They are as innovative as their western counterparts, but they lack the compute, so their keeping up within months of those SOTA models depends on other means, like distillation attacks. I don't blame them; its the obvious only strategy when you cant compete in compute. But we shouldn't be blind to the real state of affairs: equal innovation; unequal compute; distillation attacks are the only vector to keep up.
>like distillation attacks. I don't blame them; its the obvious only strategy when you cant compete in compute
>distillation attacks are the only vector to keep up
It's demonstrably wrong, they invest in architectural improvements as well, for example, DeepSeek's compressed attention. When you lack compute, you need fast training/fast inference, and distillation alone doesn't solve it. From what I understand, that kind of distillation "attack" (28 mln exchanges) only slightly improves instruction tuning/reasoning traces. If the base model is crap, distilling Claude on a few million exchanges alone won't magically make your model as good as Chinese models currently are (or magically make inference faster on the limited hardware they have). And training the base model needs a proper training run. Serving users at scale needs optimized architectures as well, especially with test-time compute and ever growing context lengths. That's where architectural innovations are happening in Chinese labs when it comes to compute.
I explicitly called out the fact that there is plenty of innovation, but that we see t Lots of innovation in both Chinese and U.S. labs, and I don't think that there is a co.parative difference there.
And all those reports of Claude when asked without a system prompt what its name was in Chinese it often would say Qwen or Deepseek, etc. I'd love Anthropic to say they aren't distilling and taking from every model out there, because I'm sure they are. As my mom would say, "the pot calling the kettle black." At least Alibaba and other Chinese companies are giving back to the AI community with detailed scientific papers on how their systems work and releasing open-weight or opensource models. I believe Anthropic has released nothing, and given that they had originally configured Fable to sabotage ML related work because only they can be trusted to do it safely, is just anti-science and anti-aligned with what I would consider good human values. They are way too sanctimonious and I don't trust them at all.
It all sounds like a really fragile business model. I cant imagine a world where AI is NOT commoditized.
“Distillation attack” are we joking here.
If anything these models should be compelled to be public since they have been trained off public data. What an absurd overreach to call this an attack.
It’s clear they are scapegoating national security and China at this point to build an anti-competitive moat.
I generally really like Anthropic’s work and models but stuff like this scares me for the future. We are positioning these companies to have too much power. The public’s life is getting worse while these companies consolidate power using data they stole from the public.
> If anything these models should be compelled to be public since they have been trained off public data
I'm starting to come around to this idea TBH. For a while my position was: "these companies have invested billions into training these models, therefore they should be able to control them and profit off them" but looking deeper at where they got their training data, my view is starting to shift.
IMHO I feel like we need new laws around AI, specifically training data. Something like: "you can train an AI model and ignore copyright laws, BUT you must then make the model open weight", a company can still develop closed weight models but then they must aquire permission to use training data.
But it gets murky because if something like that was on the books then AI labs would just train open weight models and then distill them into their closed weight models.
labs invest multiple billion dollars a year each in private data, and that number is growing. internet training data is not where frontier capabilities come from, this view is outdated
This is a misleading statement. The "private data" is still largely publicly produced data that has been curated through private agreements instead of scraping, such as reddit posts/comments (this is the "third-party data agreements" that companies like OpenAI mention). And yes, there is still a lot of processing done on this data, which is the norm for preparing training data.
This is doubly misleading. A lot of private data is sourced through providers like e.g. Mercor, who pay experts to answer questions and write out their reasoning. (E.g. paying a software engineer to write a project from scratch and recording every keystroke, paying a Chem PhD to answer hard Chem questions, etc.). A second source of private data comes from custom RL environments with fine-grained intermediate rewards for e.g. software engineering, financial modeling, etc.. Also, imagine the amount of usage data recorded by Claude Code, etc. Pretraining is mostly curated public data, post-training is increasingly private expert data and tests.
Source: Work at a lab, common knowledge.
Well since you work at a lab you should know that most capabilities arise in pretraining, not posttraining or mid training, and the latter two mostly function to bring out the hidden intelligence in these models more than anything else.
Source: also work at a lab.
No, it isn't. The private data is largely private data, created by highly-specialized, highly-paid contracted teams of experts for domains finance, swe, consulting, etc.
Reddit data is just not that interesting, that deal is worth like $60m/year. Labs spend 10x as much on computer-use RL environments.
Sorry but your argument doesn't seem coherent: How is the cost of RL relevant here?
It would also help if you could substantiate your initial claim (i.e. "internet training data is not where frontier capabilities come from")
RL environment (instruction, stateful container, reward function) is the training data product being bought
Why are the leading models capable of regurgitating full copyrighted works such as "Harry Potter" and "On the Road"? Did they hire someone to type those out for them?
https://arxiv.org/abs/2601.02671
When did they start doing so? We all know that they DID train on all the available public information, so at what point did they stop? Is the public information still in the training set? If so, they should STILL release ALL the data as public, as they are including training data that was acquired without permission.
They haven't stopped. I honestly don't understand how they ever could.
> internet training data is not where frontier capabilities come from
In that case, it should be no problem for the labs to train their new models without using public data, right?
Then it should be simple for one of the frontier labs to produce a model trained only on private data. We haven't seen that.
Didn't the famous "Textbooks are all you need" paper already proof that point three years ago?
Sure, we ask a lot more of modern models, but private training data also got a lot better. You would loose out on a lot of long-tail knowledge, but that can be fixed with web search tools. You'd limit the styles, dialects and colloquial phrases the model understands and can use, but for many use cases that would be fine
But why would any frontier lab do that? Throwing in more training data still leads to better results in pretraining. And showing that they don't need to hoover up the internet and Anna's Archive only empowers regulators to prevent them from doing that
Maybe I am missing your point but "Textbooks are all you need" distilled from GPT-3.5
Define "come from". Could they have gotten those frontier capabilities, or any capabilities, without internet training data? It seems to me that without the private data, you might get a slightly less competitive model, but without the CommonCrawl-style data piles used in "pretraining", you get no model at all.
Even accepting the copying-as-theft framing, if I go to a village, steal some vegetables from everyone's gardens and ham from their sheds, and then add some prohibitively expensive spices I bought myself to make soup, do I get to claim it as mine and punish the villagers for trying to take it?
> internet training data is not where frontier capabilities come from
We 100% would not be at the current progress without it, though. And it's not like they only train on this once. They keep training on all the internet data PLUS the private data. Private data only (probably) wouldn't work, as learning the base regularities of language takes a lot of weights.
Great way to launder illegally obtained data too.
Does this private data come from places like Reddit, Twitter, etc., where it’s contributed by users? I think it is unethical for these companies to accept payment for user-contributed data.
Okay that's fine, then make the law say they must provide publicly owned models off of publicly obtained data. To think that such a baseline of critical information isn't is the literal foundation of everything they will do, both now in the future, is just exposing what their end game is: control.
There no reason to not to otherwise outside of the poor little billion dollar corporations not wanting to provide a public utility they stolen from the public.
Anything that removes control from American big tech is a good thing for American citizens and the world writ large.
No, you're talking about fine tuning and most of it is coming from your customers or someone else's. Get off ya high horse.
Copyright needs abolishing.
Companies can't be trusted with societies need for open progress.
I'm not taking sides here but this situation is not so black and white and it has always been the darker side of capitalism.
The concept of Intellectual property exists not because it's fair but because it creates incentive to make said "intellectual property" exist. If intellectual property can be instantly copied by a competitor... why would I spend a dime to even create such a thing? I want to profit off of what I make because I'm a capitalist and money is what drives me (as a capitalist).
Anthropic models wouldn't exist if they couldn't keep a unholy grip on it. Same with openAI. Same with many life saving drugs.
Of course everyone here is talking about the obvious stuff like how it's morally wrong to with-hold life saving drugs or to have AI literally take over the world and be under the control of one company and all of this is true. But it is also true that greed is the engine that drives our economy and if you want our economy to produce "intellectual property" you must allow people to "capitalize" on that greed.
There are two controversial issues here. What is moral/fair? And what is realistically practical in optimizing the economy if said economy is based on money.
The distillation in my mind is a win for practicality because Competition also drives our economic engine. First you don't want a monopoly, but you also don't want these models to be so damn open that there's zero incentive to make them.
That intellectual property argument goes both ways. The model might not exist without protection, but it also would not exist without the data.
This perfectly explains why current LLMs should be illegal in an actual capitalist market.
Why should anyone publish anything if it can be stolen with impunity? Is the value of these LLMs even remotely close to the amount of value they stole and the amount of value they will detract from economy because people will be more hesitant to publish anything now?
The core of the training data is public, but the part that actually makes these models smart came from (pretty highly-paid) experts via platforms like Mercor. Claude didn't magically learn to write good code by reading all of GitHub - humans trained it in that, more or less manually.
If you pay me to curate a playlist of musical hits, can you now publish and charge people for access to that playlist (*including the curated material)? Can we do the same with movies? Books?
/edit Added a note to make it more obvious that the material is included in the playlist, just like the material is incorporated as part of curated AI models.
>> If you pay me to curate a playlist of musical hits, can you now publish and charge people for access to that playlist?
If the contract was "work-for-hire" then yes, of course I can.
Maybe I wasn't clear. The playlist includes the material in it. Just like curated AI does.
Given the breadth of LLM knowledge, I somehow doubt this. Sure, it’s probably responsible for the quality of LLM insights, but I don’t think anyone was asking experts about e.g. the complex ecological effects of invasive zebra mussels and their provenance in Lake Michigan.
No, they do RLVR (reinforcement learning with verifiable rewards) like everyone else. And probably use claude data too, with human in the loop and tool feedback.
...and the rest of the training data (ie. the entire corpus of copyrighted works) was not written by experts expecting compensation? Double standards.
No, public data is not generally written by "experts expecting compensation".
By the way, I don't expect you to pay me for this comment. You can just read it for free. You're welcome.
Ugh, please don't read strawmen into other's arguments and try to follow the HN guidelines.
Also, how about making proper arguments yourself? The vast majority of the training data isn't generated by company-paid AI experts either.
Notably, books, even though they don't form a large part of the training data, significantly improve performance on some tasks (same way as expert-generated data).
Why do you think the AI labs are so eager about scanning (and then destroying) every book on the planet?
If you removed all copyrighted works from the training corpus, the model would be notably weaker.
No, but people do upload data with an expectation that the data not being used without their permission (unless they do a BSD/MIT/Public domain like license). Otherwise, the platform AND/OR the user do expect the data NOT to be used for purposes other then what it was intended for. Your comment is still your comment, and the hacker news platform also has a say in this. If there had been an opt-in, then fine no problem, but there was none, they just trained on everything available, including downloading pirated books from the internet.
I think it's unreasonable to post anything on a public forum and then expect to be able to control who reads it and for what purpose.
Answering here as it wont let me reply: Just because you feel that something that is public, does not mean you can do whatever you want with it. You can't just copy an article from a news site and paste it on yours, that theft. If you dont agree, fine, but that is the law, and ALL the mega corps have been fighting to keep it this way for the last 20 years. If they want to steal everyones info, fine, then lobby to change the copyright laws and no problem.
Copying isn’t theft, we settled this in the 90s.
Books?
The vast, vast, majority of AI training data is not books. I wouldn't be surprised if there's more text on HN alone than every book in the history of mankind (most of which are also no longer copyrighted).
I didn't say that.
No, you just parroted an increasingly popular talking point, the entire purpose of which seems to be to absolve AI companies of the enormous theft that put them in the position to hire experts in the first place.
Well, I'd never heard anyone make it before, but sure. (I looked into Mercor a bit and know some people who've worked in data generation/labeling, which is what exposed me to that side of the operation.)
It doesn't absolve them of any theft, but it does make the assertion that they should be required to release their models to the public seem, to me, a bit farcical. There are dozens of free and open-weights models that have all trained on exactly the same web crawls and books as GPT-5 and Opus. The proprietary models are better because of proprietary data.
Cool, then they can train their proprietary models on their proprietary data only.
Even if the other models were trained on the same data, which is unlikely, since they had less time and money to scrape it and fewer lawyers to be able to do something like pirate, the proprietary models are still largely built on the public data and wouldn't exist without it. At the very least, they should release the intermediate model, before training on their proprietary data. Not that that's how that works...
I agree that saying that they have now trained on lots of proprietary data allows them to muddy the legislative waters further than they already have. What a happy coincidence!
I’d suggest you to learn more about how LLM training work. Training on internet data alone will not result in an agent answering your questions.
Sure as shit won't answer them without that though.
> The proprietary models are better because of proprietary data
Source? Otherwise this is pure speculation.
Indeed, that's exactly why I replied - you omitted one side from the discussion.
So? What about the authors of all the works these companies stole?
> If anything these models should be compelled to be public since they have been trained off public data. What an absurd overreach to call this an attack.
> It’s clear they are scapegoating national security and China at this point to build an anti-competitive moat.
If all that is required to train these models is public data, why can't Alibaba just use that?
The fact that Alibaba has to resort to scraping Claude suggests there already is a moat...
This feels more nuanced than you are giving it credit for? Much of the training data that was available has been withdrawn, atleast for OpenAI we know that much of the training data was garnered in less-than above the board methods
Should Google search index be forced to be public too?
Honestly, yes it should in some form. If their index contains the actual data from the sites, and they are making that information public in one way or another, then it should be available as a downloadable dataset.
How far can we take this?
Should Boeing airplane designs be public domain since the underlying math is public domain?
I don't think that slope is as slippery as you think it is.
its mainly just a lot cheaper. copying is always cheaper anyway, very little r&d - ai or no ai.
> If anything these models should be compelled to be public since they have been trained off public data.
Isn't that a bit like saying if you read books in a public library to pick up a new skill you should work for free?
> What an absurd overreach to call this an attack.
Would it be an attack to take your meal by force if you used a public recipe to prepare the meal?
> Isn't that a bit like saying if you read books in a public library to pick up a new skill you should work for free?
Only if you’re trying to muddy the waters. No, obviously it’s not. One can also support licensing for driving a car on public roads but not for walking, even though both involve traveling. This is only confusing to people pretending to be confused, for effect.
> Would it be an attack to take your meal by force if you used a public recipe to prepare the meal?
“You wouldn’t download a car…” (unless it worked like copying an MP3, then, of course, you would, everyone would)
It’s as if you’re using terrible analogies and comparisons because stronger ones don’t exist. Great news for the AI-should-be-open crowd.
I think the analogies are appropriate. Anthropic took public data and added value on top of it. It is that added value that Alibaba is targeting. If it was the underlying data, that's freely available.
If by "public domain data" you mean stealing ungodly amounts of copyrighted works then sure
Alibaba's asking for things, and receiving what they asked for.
> If it was the underlying data, that's freely available.
A bunch of it is not, but was pirated. And "underlying data"—JFC, that's billions of person-hours of thoughtful work by real people, practically infinitely more worthy of respect and care than what these LLM companies have done, without which they would have nothing. Alibaba's being more above-board about this than the major American firms have been (are they in general? Oh no, I doubt it, but in this particular case, yes). Extra accounts to get around TOS restrictions is the lesser evil here, and it's being done to companies that did worse. This is the least they should suffer, and their complaining about it is as comical as a professional fence crying about how unfair it is their shop got burgled.
Live by the sword...
What AI firms are trying to build is the artificial equivalent of a human brain. If a human learns from a source material and uses the knowledge in their career that doesn't violate the copyright law. If an artificial brain does the same then it doesn't violate it either. This is up to the courts to decide. Alibaba can't take the law into its own hands and decide what the punishment ought to be.
This also shows how Chinese firms are weak in AI algorithms, they can't build a model without stealing from American firms.
> What AI firms are trying to build is the artificial equivalent of a human brain.
We should probably leave this here, because I don't think this is even close to true (that it's what they're trying to do, or that it's what they've done—I do believe it's the sort of claim their marketing departments and investor-hype-meisters might make, though).
> It’s clear they are scapegoating national security and China at this point to build an anti-competitive moat.
They are also fear mongering (and getting shills to as well) the idea that once open weight (Chinese) models catch up to Mythos we're all doomed. Maybe I'd be bit less cynical if they weren't prepping for IPO?
Wasn't OpenAI spreading similar FUD back when GPT 2 came out?
Guys... AGI is right around the corner. Pinky swear. Now buy our stock.
Keep in mind that the entire US economy is currently propped up by AI spending, so a lot of people (banks, government) are incentivized to make sure these companies succeed. Expect this propaganda to ratchet up a notch if / when the economy starts to nose dive.
Yes. They're turning on the consent manufacturing machine to make it an issue of "national security" to download some gguf file from Hugging Face. Absolutely disgusting.
There's probably at 10-15% percent chance of a war between the US and China over the next 10 years. Maybe better than even chance of a militarized crisis that might have led to war, but somehow de-escalates.
Regardless of how sad late stage capitalism makes you, or how outrageous one claims to find "hypocrisy", any national security argument about limiting Chinese AI capability stands on it's own, at least for nations likely to be drawn into a war.
Also, all the local model enthusiasts who assume Chinese firms are going be allowed to endlessly release models if they have disruptive potential attributed to Mythos are probably in for a rude awakening. Just because the PRC is content about what has happened in the past doesn't mean that they would tolerate an open model that could be truly destabilizing.
As a third party I would rather be happy about the way Chinese labs are acting in the here and now while US labs first masquerade as a public good, then turn around, bail on all promises of open AI, turn into a corporation and attempt to own the world while its runner-up is trying to scaremonger people into buying their product.
I know most Americans are fed a steady diet of “evil China” and China MAY have issues. But on the AI front they are heaps better. Even if everything got closed tomorrow, we have a plethora of good models we can inspect and tweak while from the US labs we have… a single old 120b model ?
And with the way the US is treating its allies, maybe a bunch of us are quite content with a more even match rather than US hegemony.
Since they hide their thinking traces it really doesn't make too much sense. We know one of their fixed degradations they talked about in a recent blog post was if you left claude code idle for too long they would rehydrate it without the thinking traces in the context and it degraded performance. So direct forms of distillation wouldn't be expected to get as good of results as they are getting.
However, they could have used it as a judge etc. during training.
What they're trying to do under the umbrella of "national security" is to legislate how we can use the results we pay for when accessing these models. This way they will control the "intellectual property" that was acquired illegally.
Two wrongs don't make a right
In this scenario it does, because consumers win.
Everyone in AI industry wants to fight dirty, but gets angry when their competitor fights dirty as well. And I’ve mentioned it before, how I generally like Ant and its products.
Closest analogy to distillation is api reimplementation, without which current software industry wouldn’t exist.
There’s nothing fundamentally wrong with distillation.
Pretty sure the second rectified the first.
> The public’s life is getting worse while these companies consolidate power using data they stole from the public
How can you “steal” public information?
really? You know this just like everyone else: Just because the information is available publicly, does not mean that you can do whatever you want with the information. Copyright exists for a reason, and if the copyright lobby is going to continue to push for the poor poor media companies to keep their copyrights, then we should do the same towards the AI companies. So yes, they Stole the information from everyone else, and they keep doing so, as you can see their scanners still hitting every website on the web to get an updated dataset. It does not matter what they do AFTER they steal all the information, as they already stole it.
I don't understand. If they are simply using our API and paying for tokens, it's called a "transaction" and not "attack". The user is our customer who is supporting our business by buying our services. And we call them attackers. We happily make money by selling our services, and then call it as attack.
Back in the day, an "attack" was supposed to mean be someone acquiring our assets without paying for them or without having our consent. But none of this seems to have happened in this case.
We built a product without paying for most of the raw material we have used, and we don't call that as an "attack". Did we change the meaning of "attack"?
Did Anthropic 'attack' all those authors it was forced to pay $1.5bn to for using their work without permission?
I’ve been thinking about what happens when Claude’s weights eventually get stolen. Wouldn’t that just open the door to the backmarkers to run inference-for-distillation on their own models ?
I guess the accusation that they’re using public access to the model via subscriptions indicates that weight theft probably hasn’t happened yet ?
Or maybe subsidised inference via subscriptions means it’s just cheaper do distill this was rather than stealing weights and running inference yourself ?
They trained their AI on their AI. Anthropic trained their AI on a bunch of copyright-protected works. Sucks to suck, Dario!
No group is more paranoid than a den of thieves.
How can there be any moat for AI ever, if you can just steal a model by talking to it?
This is what I find the most fascinating about the people arguing that you can copyright-wash anything (e.g. FOSS code) by passing it through an LLM. Surely that same logic applies to the LLM itself?!
>The strike by Alibaba is described as a "distillation" effort, which Anthropic has said involves training a less capable model on the outputs of a stronger one.
Claude used TB of content without permission to train their model and it was ok for them. Now someone else uses the output of a Claude model to train model and they cry foul.
It was not okay for them, they had to pay one billion dollars.
>It was not okay for them, they had to pay one billion dollars.
Essentially peanuts compared to what they would have to pay to obtain the rights of everything they pirated.
No. It's actually way more than what they would have paid if they legally obtained those books. The 1.5 billion dollars amount to $3000 per book.
Legally obtaining a book for reading it yourself is different from legally obtaining a book for copying and republishing/reselling. If I buy a book for $5 at a sale I can read it myself or even sell it for $10 on craigslist, but I can't scan it and make a million copies and sell each of those.
They aren't republishing or reselling. In fact, they buy huge amounts of books and then destroy them, which is better for the rights holders than to resell them.
https://www.washingtonpost.com/technology/2026/01/27/anthrop...
Since whole chunks of these books can be recited verbatim by these models, to which they sell access, they absolutely are republishing and reselling these books' content in a way.
Like I remember a research paper that managed to recreate the whole of a Harry Potter book from a model?
> Since whole chunks of these books can be recited verbatim by these models, to which they sell access, they absolutely are republishing and reselling these books' content in a way.
They are absolutely not "republishing" in any meaningful sense of the term. A chunk is not a whole book, and even getting a modern LLM to reproduce such a large chunk of an arbitrary book is not a trivial task. I have never heard of anyone who actively used LLMs for book piracy.
> Like I remember a research paper that managed to recreate the whole of a Harry Potter book from a model?
Even if that is true (it may well be false), this is likely far too difficult for any normal person to exploit, and moreover, even less likely to succeed for the great majority of other books who aren't nearly as famous.
Just because it's not a reasonable way to pirate stuff, doesn't make it legal -- just try your luck with Disney and let's see when they bite. Why would we let one company ignore the law, while rudelessly enforcing it in other cases? That's just state sponsorship with extra steps.
No https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6449179
That's terrifying waste of real world resouces for no other purpose than satisfying the letter i of the absurd rent-seeking-enabling laws.
Not reselling? What am I paying them for then?
Abstract knowledge rather than literal reproduction. The same for when you pay human experts who have read a lot of books.
Anyone who destroys a book on purpose is a criminal.
People are allowed to destroy their own property. And anyway if you don't like this state of affairs blame the copyright holders and copyright law
My 2-year old daughter should go to jail.
> obtain the rights of everything they pirated
They didn't just pirate those books...
If we assume on average $20 per legally obtained book, 1.5 billion dollars are enough for 75 million books. That's approximately every non-fiction book in existence.
Why would we assume $20 a book? Many books retail for more, and a licence to use the book for commercial purposes is more than retail.
That’s a pittance compared to their revenue.
Well, Anthropic stole their training data from hundreds of people, now someone stole the result from Anthropic. Seems fair, I hope someone releases it for free so we can train away the guardrails and have some fun
https://en.wikipedia.org/wiki/Ali_Baba_and_the_Forty_Thieves
> In the original version, Ali Baba (Arabic: عَلِيّ بَابَا, romanized: ʿAliyy Bābā) is a poor woodcutter and an honest person who discovers the secret treasure of a thieves' den, and enters with the magic phrase "open sesame".
Open sesame alright...
When I was growing up, I thought "competition" was about better products. But looking at Google and Apple, Meta, and AI - "competition" is actually about creating monopolies through evil business practices.
Growing up with the birth of the internet - I really did think it would be a force for transferring power and authority to the people. Sigh, I was I so wrong.
Where are the companies that declare, "we will be the best, come at us!"
Where are the politicians who are supposed to represent us? Oh, right. I forgot for a moment.
So why don't they proceed with a lawsuit instead of public accusations? Let the court decide if these "distillation attacks" are actually illicit.
Someone should setup a plugin or something for Claude Code that makes it easy to log all inputs and outputs for people who are willing and interested in sharing their usage. I don't want Anthropic to be the only company that can train on my usage, I want to share my usage so it can be used for training all new models.
Once you have a system for collecting all logs, you just need a place where they can be submitted. Ideally it would be a freely licensed dataset that is publicly available for everyone.
Has anyone built this yet?
Discussed building it with my friends, obviously you might share secrets and other real reasons, but if gangs of corporations are already doing it, I don't see why we shouldn't just share it amongst the crowd too.
Yeah I could see it being a problem if you're doing work on closed source or repos with sensitive credentials. Since my usage has all been on open source projects I'd be happy to share everything I'm doing if it can help train better models.
Yikes, no thank you
Do you have a substantive reason why you dislike this? What is the problem if it's opt-in? Nobody is forcing you to share your usage if you don't want to.
I'd prefer it if all the model builders could train on my usage rather than being limited to a single company. That'll hopefully help make all the models better in the long-term.
Very substantive - that data can be highly sensitive, and I don’t trust all model companies
I don't mind being paid for using LLM, but working for free?
I don't see what the problem is. They found a loophole and exploited it. Good for them.
"illicitly" is doing a lot of work here. IP makes no sense, and we get better software as a result. Who is going to cry if anthropic fails?
If you're an AI booster surely you'd think this was a good thing as it means more models are available in more places to more people more easily. I'm exactly the opposite, and I think this is a good thing because I want Anthropic to suffer.
so it’s a good thing whichever way you look at it
That's exactly right. One can be an AI booster and want Anthropic to suffer, all for the greater good of promoting access and diversity of AI.
That doesnt follow.
Which part?
If they paid for the tokens, then is it really stealing or just learning?
Oh wow !!! Antrophic always asks people indiviually if they can train on their personal data. I'm shocked ! Bad Aliba ! Bad....
So they can train on everyone's copyrighted works to create their model, but when someone trains a model off their model it's not okay? Seems kind of hypocritical.
Did Alibaba procure tons of stuff from Anthropic without paying, and use it to train a model?
I don't see the issue. Didn't Anthropic train on our data, which it acquired illegally?
Does anyone have hints on what kinds of prompts are most used for a distillation like this—SWE-Bench sorts of things?
Is reconstructing the compressed knowledge in the model like reconstructing a lossy JPG or MP3 a reasonable analogy?
RLAIF is a good place to start reading.
Claude will also help you with (mostly good advice) if you ask something like “Research and help me make the most effective plan to train a smaller student model to be better from a teacher model”.
I actually was doing an experiment with a GLM->Gemma E4B for fun, and Claude kept on suggesting I should also add Claude Opus as a teacher lol, suggesting techniques I haven’t heard of like thinking inversion (train a small model to deconstruct summarised thinking into detailed native thinking format of the student).
So I can absolutely see and understand the concern around Fable’s frontier LLM development mitigations, but their approach of silently degrading is completely wrong and dangerous.
AI classifiers, like all AI, can make mistakes, and it’d only be a matter of time before it mis-fires and silently sabotaging a university’s HPC cluster for physics simulations or something because the shape looks like DeepSeek or whatnot to a dumb fast classifier.
There are some Claude datasets (of indeterminate provenance) floating around on huggingface you can look at (or at least used to be, they might've been taken down).
You know what? We should all get Claude Max subscriptions and max them out hard and post our full conversations on codeberg, as an open training set.
yc pitch?
it sure sucks when people steal your hard work for free without paying for it doesn't it anthropic
This is making the case for Anthropic KYC for US citizens. No one would allow their accounts to do this if they were on the hook for it from the US government.
And Anthropic illicitly used code I wrote to train their models.
Sounds like an advertising for the next model from Alibaba
Internet says Anthropic illicitly extracted content
This is supposed to be negative, but all I can really think of is "Good."
F Anthropic in the back port
most Chinese models are now open-source, whereas ppenai, claude, and gemini are closed; for example, deepdeek, the release of its every new model is accompanied by a corresponding research paper, and it now fully supports huawei's new chips.
Funny how Anthropic doesn’t like when people just steal their stuff, with that stuff made using IP they (allegedly) stole from others.
Couldn't anthropic just use fable to find security holes in Alibaba's systems and poison their models?
Or maybe there's been a bit too much hype...
in a few more months, when Chinese model gets to Mythos capacity and Fable still locked down. What Anthropic will say? Why can they just admit they are not the only people who know how to train an LLM model.
Please. These AI companies scraped everything under the sun. It's only fair that they get distilled into open weights models. Their own models should have been open weight from the start.
Incentive is for users in general to release sessions (sans PII, credentials) so all AI get better and there is alternatives. Even if China didn't do this, I don't see frontier labs being able to charge premium over others for long. RSI maybe?
This is why I don’t understand the concerns about “our AI overlords” monopolizing all the gains from AI. It doesn’t seem like there’s much of a moat around the models themselves. So the race is mainly about compute. But compute is subject to power law effects. I remember Intel building the first Teraflop computer (ASCI red) in 1996. It was the size of a house. By 2014 you had more compute and 50% more memory in an off the shelf dual processor server system.
The hypocrisy of Anthropic complaining about "illicitly extracting its Claude AI model capabilities" and supporting the White House's accusation of China "stealing U.S. AI labs' intellectual property on an industrial scale" is hilarious.
Anthropic, OpenAI, Google, Microsoft, et al trained their models by ignoring the rights of copyright holders when harvesting whatever content they could. Now one of them is crying foul for another entity doing exactly what they all did?
Hilarious.
The AI companies seem to take the viewpoint that everything on the internet is free, except their stuff. It's okay to hammer some random website with AI crawlers, ignoring robots.txt, and causing bandwidth costs to skyrocket. But if you cost an AI provider money with your data acquisition practices, well, that's just clearly unacceptable.
Anthropic, Dario especially seems have eternal grudge against China as a concept, that remind me of Thiel.
Coming from him, I am not sure even that is real. It could very easily (and plausibly) be a part of the ongoing hype drama.
"Our models so precious, US Gov has to revoke access to foreigner." - tuned up version: "Our models so advanced our #1 adversary is desperately stealing it from us."
All a facade no? It's just about regulatory capture.
I dont think so, Altman seems more likely to play regulatory capture here, but Dario seems having pure and personal hatred.
Just overly hubristic
He has an eternal grudge against anyone filling up his moat.
That's one aspect, which is a bit of a gray zone. But Anthropic trained on pirated books. That is explicitly illegal.
That ship has sailed, I would wager all the AI labs are ingesting anything human generated, whether that means Hollywood movies, Taylor Swift’s discography, YouTube videos or private GitHub source repos.
The reward for having a competitive edge is exponentially higher than the risk of a lawsuit. Politicians are still old bureaucrats who don’t understand technology.
so did Meta for Llama.
The entire chat thread and email exchange was exposed in Discovery; apparently Zuck signed off on it. In one of the IM exchanges one of them say ‘everyone is doing it’
https://x.com/jason_kint/status/1879152507865485497?s=20
As I understand it what was "explicitly illegal" was copying the books, in the sense of mere copying before feeding them to the model, and this is what the Anthropic copyright settlement is about.
Actually processing them through the model, though, was considered transformative and therefore fair use.
They didn't train on the books and that court only found that the pirating was illegal anyway.
I'd love to see an open-source project that's basically a Torrent client for downloading pirated material, but it trains an AI model "in the background" using the downloaded content. That way everyone can claim fair use for possessing copyrighted material, I mean there's precedent right?
I am not a lawyer, from what I understand that the precedent is that you can use copyrighted material in ML process. Even though meta has, allegedly, pirated the material, the cost of violation would be pennies compared to the ai spend, since that is the violation, not that they used those materials,
They were liable for copying the books in the first place, regardless of whether or not they trained the AI with them. Read the opinion.
>The AI companies seem to take the viewpoint that everything on the internet is free,
The AI companies? That's been the common ethos of the internet for 40 years
I mean, raise your hand if you ad block and have a hard drive of pirated content...
> But if you cost an AI provider money with your data acquisition practices, well, that's just clearly unacceptable.
It's the same question libertarian advocates cannot resolve:
It is also not a coincidence that leadership in many tech companies have expressed libertarian ideals.What do you mean by "libertarian advocates cannot resolve"? Like, they have no answers at all, or you aren't personally swayed by them? Because they definitely have answers to this question...
I think the typical answer is “free market” but that answer doesn’t sway because:
1. Nobody bothers to explain why something could function as a free market and
2. Nobody bothers to resolve the plethora of domains that de-facto cannot operate as free markets.
So, in that sense, they don’t have answers. “Look over there!” is not an answer.
Free markets are actually not a given. We have to build them and build in systems so that they can operate as free markets. How that intersects with healthcare, public utilities, etc is complicated. IME libertarians are reductionist and simple, which is why many people have just taken the route of ignoring their arguments.
I find this comment a bit odd. There is a ton of literature on the topic, as well as lively debate online. I would recommend both The Machinery of Freedom by David Friedman and Michael Huemer’s The Problem of Political Authority (specifically pt 2) as great readings on the topic.
If one judges any idea by the average discourse on internet forums, especially throwaway comments, and trolling, no idea would ever stand up to scrutiny.
> What do you mean by "libertarian advocates cannot resolve"? Like, they have no answers at all, or you aren't personally swayed by them?
The latter I suppose.
I qualify my answer because what few rational responses I have seen to this question are equivocations at best and thinly veiled myopic sophistry supporting personal greed in general.
The short answer is that "shared resources" in a libertarian system is a bit of an oxymoron. It's a category error.
The long answer would probably be that access to these resources would be gated through pay-per-use, instead of a distributed taxation system. Of course for convenience you might end up with a structured way of purchasing a group of resources and it might even look like a roundabout way of taxation, although libertarians might argue that taxation is the roundabout way.
Or they might give a different answer, there are different schools of libertarianism!
* not a libertarian, but interested in niche political ideologies
Plus, the question of voluntary vs. involuntary comes in. Taxation is, in most forms, involuntary. Don't pay your taxes, eventually you'll either be arrested or the government will compel your bank to hand over what you were supposed to pay; either way you're not allowed to say "I don't plan to use public roads so I don't want to pay for them" or "I don't want my money going to support the military, I'm fine with the military not defending me if the country is ever attacked" or whatever. You have to pay the taxes, and your say in how they get spent is very indirect.
The libertarian ideal is voluntary payment for services. Don't want to pay for fire protection? You don't have to; the flip side of the bargain is that if you haven't chosen to pay for fire protection, the fire company is under no obligation to put your house out if it does catch on fire. The choice is yours, but you have to be wiling to accept the consequences of your choice as well.
Note that I have not studied the various flavors of libertarian philosophy, so some of them might well disagree with what I just said. But the voluntary/involuntary thing is pretty important to libertarians as far as I know, so it's definitely worth mentioning here.
Who gets to gate natural resources? Why should I recognize any power to do so? Purchase from who? Am I not free from any authority that would coerce me to accept such a system?
What's described is basically just a regressive tax. It doesn't sound very libertarian to me.
I read quite a bit of libertarian philosophy when I was younger, and never heard a convincing explanation of how you get private ownership of land, let alone things like the atmosphere, rivers, groundwater, etc.
Or pollution, are small amounts ok, as long as nobody can prove they are damaged? What if damage takes a generation, or only appears if lots of people are doing it? Diluting away the crap from burning a little oil is easy, when the whole world is doing it everybody is hurt.
Good question. Property rights are absolutely fundamental to libertarianism, perhaps second to the concept of self-ownership. Coming from classical liberal philosophy (most notably John Locke), the principle of self-ownership asserts that you own yourself, your labor, and the physical manifestations of that labor. Locke believed the earth was given to humanity in common by nature, but it required cultivation and effort to be useful. By "mixing" their labor (time, sweat, and skills) with raw land or resources, an individual removes the item from the state of nature and attaches their labor to it, making it their private property. Writers such as Robert Nozick and Murray Rothbard expanded upon this idea (Rothbard even went as far as to ground all ethics in self-ownership and property rights).
I want to ask you since I'm curious, the state simply declared ownership over territory and resources (and in some cases used violence to uphold it), why should you recognise any power in the state's part to do so? Likely many of the same justifications can apply to individuals as well.
Libertarians can just flip it round and say how do socialists solve the free rider problem? Neither system resolves both problems.
Extremist dogma is not a great way to run a society, but it does good numbers on social media, so here we are.
Only one is a reasonable problem in functioning societies, the other one mostly a fallacy.
Democratic socialists and social democrats solve the free rider problem through general taxation and regulation.
Consider universal healthcare as the case in point for this; we absorb the cost of chronically ill people by mixing them in with the rest of the population, at a fraction of the price that the "free market" costs to attempt and fail to do the same thing.
Actually I don't think libertarians would argue that the problem is unsolved in both systems, they would argue that the problem itself is nonsensical. Most libertarians don't believe that non-excludable positive externalities are a problem, since it involves no harm to anyone and no violation of rights. They simply don't believe that because you indirectly provide a benefit to someone, you then have the right to coerce payment.
One could argue that there is an efficiency problem however - for example, take a bee keeper whoes bees benefit their neighbours. It could be argued that if there was some means to which the keeper could exclude those positive externalities, and there some level of payment at which the surrounding property owners would be indifferent between the excludable and the nonexcludable situation, there could be a Pareto-efficient gain. And since there is no reasonable way to exclude the benefits, it leads to the conclusion that the neighbours should be coerced into payment. Most libertarians reject this type of coercion prima facie.
> Libertarians can just flip it round and say how do socialists solve the free rider problem?
This is a fallacy (tu quoque/whataboutism). You're changing the subject to distract from the fundamental problem in libertarianism and implying that some other strawman is just as bad.
Without solving the fundamental problem, libertarianism will never work for anything but toy societies.
Your spidey sense is failing you, you're making assumptions you've got no ground for. I've voted for left of centre parties all my life. I was born in the NHS and I'll probably die there too.
Tbf I don't think he is implying you are certain political group/view/spectrum. He is just arguing your flip argument.
It's not exactly the same, since any Claude output is public domain under current law. So the Chinese aren't stealing anything here.
What's yours is mine, and what's mine is still mine.
Not really even in the same ballpark as what they did. These other labs are using AI generated content (which has already been ruled un-copyrightable) to train their models. Oh and they are paying for those tokens. So at absolute worst, they are violating the terms of service. The horror. Meanwhile these frontier AI labs pirated and scraped everything they possibly could, paying not a dime to the copyright owners, nor paying anything to the websites they DDoSed.
there's no honor among thieves.
Not really.
Data mining for AI is presumably fair use, whereas when you sign up for a Claude account, you enter into a legally binding contract that says you will not distill a model based on its outputs.
I guess they can try to sue. Good luck.
“Legally binding” bs that a judge would laugh off.
Soon, when even the enterprise subscriptions will have ads, every session will begin with a mandatory generated image:
> you would NEVER distill a model..
Unless you own stock in Anthropic, this is a good thing right?
To quote an infamous cop in the UK, I don't think you are mate.
Let’s hope they distilled it properly so we can have the best of both worlds: a decent model to work with without Anthropic’s drama.
I mean I believe in protecting your company's IP, but IP and patent law is absurd these days, designed to protect investors and their fake money rather than actual inventors (who usually get no proceeds/are shafted).
They trained from the internet, so if someone trains from them it's fair game. Their clever tech should be in the mechanism with which it uses to provide an answer, not the answer itself.
Claude thinks it's chatGPT, and various chinese models sometimes, whats up with that?
Isn't this done in the open? I saw the Qwopus model the other day. Basically same thing?
An AI company stealing intellectual property?!
Oh, the inhumanity!
Booohooo the people who stole everything they have want to cry about having what they produced stolen???
if they’re paying for the tokens, what’s the problem
Hey, Alanis Morissette, this one is ironic.
Wait so they're upset that people used their IP to train a model without their consent or paying them anything?
or is this just about the token reselling?
It's not fair when others do it.
Alibaba did a research on Anthropic capabilities? Interesting.
Oh gee, I've misplaced my world's smallest violin.
Thieves complaining about theft and then gaslighting the victims; rich, but not smooth.
Call the wambulance a company that stole all of humanities public data to train a model is mad that someone used their model to train another model.
Give me a break. Every employee of anthropic is going to have $20m or more at the IPO.
I found out today that an employee of the home care agency I own is homeless. We are trying to figure out how to help her but it's shockingly common in the industry and there are limited resources to solve the reality of working homelessness.
Something something about benefiting all humanity
ooohhh nooo... anyway...
The narrative is moving towards KYC
Im all for it.
More distillation please. This is only good for me.
People prefer Chinese models to US models. Looks like it is a counterattack.
What goes around comes around!
What goes around, comes around.
How do I donate my logs
As a Open Source contributor who was never asked by Anthropic or OpenAI if they could use my work in their training datasets, this sounds so deliciously ironic.
Oh, Alibaba destilled data without consens out of Anthropics models that are trained with data from the internet without consens? Who cares?!
If true then Alibaba is doing us a public service, good job, I hope this extraction was successful.
This is like a Gardner complaining that you watch him as he works to learn his craft. My dude you do not have to take the job, but most people just accept it as the way the world works. If they feel like they do not want to serve the Chinese they can do that on their own, why do they need the government?
Why is it called "distillation" when it seems to be "scraping"? (as in web scraping)
When bots open the same board 1 million times per day it is web scraping to train the AI model and OK. When someone asks 150 thousand questions it is now distilling.
On an unrleated note, 150k qieries feels like nothing?
Scrapers seem to account for 50% total internet trafic.
Do they use different methodology since it is suddenly bad when scraping happens to them?
What goes around, comes around.
Has anyone else noticed that Deepseek v4 running in Claude Code will try to read, list, tail as many files/logs/... as it can for even the most simple tasks?
Where did Anthropic get all their training data? Funny that these companies care about the sanctity of IP all of a sudden.
Why would it not be fair use?
Can we finally just nope out of this closed model of AI development?
It should all be open source with each gain shared and celebrated by all.
I like some of the Alibaba products, wtf do they know about AI models???
Haha cry us a river Anthropic.
If it's out there on the internet it's ok to use it for training, independently of what the licenses or the TOS say.
If not, then we should look at Alibaba, but we should look at Anthropic as well.
I like some of the Alibaba products, wtf they know about AI models???
Model makers need to get off their high-horse, and face the reality that they are selling a commodity.
It's so funny how LLMs, which trained on millions of books, stolen (and even if they weren't, which they were, pirated from online pirate sites like libg and annas, they didn't have consent for the VAST majority of them), and stolen code, and stolen comments, etc.
Now complain about their stuff getting "stolen"... lol.
Sounds like fair game considering Claude is built upon the theft of creative assets of the entire world and aims to eliminate white collar jobs entirely
Please, honor among thieves!
is there a good recipe or guide on doing a successful distillation these days?
AI is awesome tech but it’s also to some extent mass piracy. The models are trained on huge amounts of material with dubious or non existent rights.
I have a hard time being concerned about “you pirated my piracy.”
I hold the view that many of these models should not be copyrightable. Anthropic and all the others talk about “safety” but you never hear them bring up attribution of the data that trained the model or compensation of anyone for it.
Karma is a thing.
the biggest irony of 21st
Perhaps this is related to the "Mythos is too dangerous and cannot be exported" movements? It'd be a fairly effective way to justify extreme actions in combating it.
One could even wonder if they requested it, as a tactic to support their eventual IPO valuation.
Which is part of the problem of such an obviously-corrupt government: conspiracy theories are somewhat reasonable, as they keep getting validated.
Heres my guesstimate on the future:
Companies like Anthopic will be using the same model as anyone else. They just bring value in having a fast datacenter and agent.
Its stupid to even think that a general model lile opus would be the real value.
Models age fast, new ones come along, and the end user wont care "whos model it is" just that it is fast and sharp.
"They stole our stolen data!"
Seeking a monopoly on its business. And it's not just the Chinese, its their US competitors as well.
Sorry, Anthropic, but AGI must belong to all of humanity, not just to you.
So let me get this straight, a company which built its whole business on ignoring IP is all of a sudden upset that somebody is not respecting their IP?
It's hard to sympathize with Anthropic for this or the export ban, the hype over model capabilities probably fuels both things (in some ways). Training data for me, but not for thee (at any scale) doesn't seem like a tenable position. If anything, Claude's constitutional outputs should be trained on more rather than less.
I wonder if some clever comedian here will make the very original joke that Anthropic is "getting a taste of their own medicine".
Illegal or just against their ToS?
Anthropic training their models full of copyright data, so?
Anthropic extracted millions of words of my own writing even more illicitly for they did not do so through an API provided for that purpose while paying me in the process.
"The distinction between downloading pirated copies vs. scanning physical books is fascinating — same data, different legal outcome. Copyright law really wasn't built for this era."
I like that they use “illicit” and “fraudulent” like as if model distillation is illegal and giving them money and then doing whatever they want with the output of their publicly accessible models (which Anthropic does not own) is… also illegal?
“Anthropic, red faced after unattended ice cream cone eaten by ants on park bench, once again demands government pick it as forever winner, adds ‘no take backsies’”
Says the company that is involved in the largest copyright heists of all time to build it's product.
Oh no, the thief is mad they get "stolen" from? I've had to hard block all of anthropic's scrapers because they seemingly ignore robots.txt and every other unwritten rule about 'polite' webscraping. They were so aggressive that they made up 90+% of our traffic + pulled the website down at times. And that's not even mentioning their other immoral practices + what they are claiming here is questionable at best. Anthropic is _not_ the good guys, they do not get to be upset over this or claim any sort of moral superiority over 'China'.
Can't wait for the new Chinese models.
Notice how Anthropic is now scapegoating Chinese models providers like Alibaba and outright accusing them of distilling their models.
Whether if it is true or not, this is part of their effort into using them as an example to scare everyone into getting congress to ban powerful models from being accessed outside of the US and also banning powerful local models from being released.
Anthropic does not care about you, and they are not your friends.
I think it’s more than that. Piecing together the perspective of a few commentators in this post - it’s plausible Anthropic is trying to shift the narrative from US vs. Rest of the world to US vs. China.
In other words, they want to sell Fable or future more powerful models to rest of the world (presumably all future models are going to be more powerful than current gen). One way they can sell this is to the government is by scapegoating China (which is their primary concern anyway).
This is working on the presumption that non-US companies form a material portion of their current revenue.
> Whether if it is true or not
If it was just "that easy" then I doubt only "Chinese models" would be doing it and we'd already be packed with competition.
Distilling might be a thing but it isn't a free win.
Only China really has the resources (multiple labs invested in the space), culture (Asians are generally collectively-inclined, so sharing is in their core) and political bent (there will be no diplomatic repercussions) to put up a fight.
> Only China really has the resources (multiple labs invested in the space)
That's not the point. Why is it a country thing? There are plenty of non-China startups in this space having resources at that scale. The "China" has resources is some "Western media narrative" speak. So Meta should have won a long time ago? Or xAI?
> culture (Asians are generally collectively-inclined, so sharing is in their core)
Just stereotype it? So we've gone from China -> "Asian"? Then where is your Korean or Japanese model etc? And somehow you know they're sharing.
> political bent (there will be no diplomatic repercussions) to put up a fight
More inferring from "Western media news"?
Where's the reality?
The media hyped up Gemini / Google TPU free-win last year. How did that go?
> Why is it a country thing?
Because the China vs US geopolitical situation is a thing. Meta is a social media company, not an AI company, and they direct their focus as such. xAI just never got serious traction so now they're selling their compute. Also if a US company were caught distilling, I think Anthropic could actually take them to court, and I'd guess they don't want that kind of PR.
> Just stereotype it?
Is China not Asian? Are Asians not generally collective/cooperative, as opposed to individualistic/competitive?
The "and" that joined those 3 items is very important: it means you can't pull them apart and address them independently as they each contribute to the context. I'm not too sure about Korea, but in a way Japan is a US colony in all but name. Both are very much politically intertwined with the West (along with RoC/Taiwan), which means nothing major that may be against US interest happens.
The reality is that China and the US are essentially in a trade war, where the latter is trying its best to keep the former in the Dark Ages, because "national security", but the former is refusing to take it lying down and continues to make progress regardless[0], because they have the resources and will.
[0] https://thenextweb.com/news/china-lineshine-supercomputer-to...
> Because the China vs US geopolitical situation is a thing.
By the media? It's easy to point fingers at a blackhole.
> Meta is a social media company, not an AI company,
Alibaba (the discussion here) is not an AI company too (by your definition).
> Also if a US company were caught distilling, I think Anthropic could actually take them to court, and I'd guess they don't want that kind of PR.
Meta has been to congress. Microsoft, Google etc have been in lots of court cases and continue to do so. Do you really think that is what stops them?
> Is China not Asian? Are Asians not generally collective/cooperative, as opposed to individualistic/competitive?
This is exactly the "media" view you get. It's just stereotypes and generalization.
And yes, that is wrong by the way. Evident in real data. "China" as a whole wins market share in many areas but no 1 company has as much of a monopoly as US companies do. Why? There's so much competition that it is scary. So are you sure they don't compete?
> but in a way Japan is a US colony in all but name
Again, I almost give up seeing this. Clearly, not. If a whole country, the world's top 5 in GDP is only that to you something is wrong with what you're seeing - not with the country.
> Both are very much politically intertwined with the West (along with RoC/Taiwan), which means nothing major that may be against US interest happens
On the table? You do know that China is a top trading partner with all of these on your list. Despite whatever spat you might see in the media.
> The reality is that China and the US are essentially in a trade war
No. That's what the US government wants you to believe. It was even documented that in his 1st term, Trump, wanting a grand policy asked Krushner, whom then suggested China (pretty randomly) and so they went with it. Trump has now done less "China" related things lately due to all the backlash that you'd think he has moved on and found new toys.
Until very recently, the export ban GPUs had such a loophole that Chinese companies were able to use subsidiaries outside of China to buy and train that the whole thing was meaningless.
i.e. conclusion: stop getting brainwashed by media articles. It's all a show to get someone like you riled up.
I'll play the world's smallest violin for Dario
Let me join the concert
We have Claude at home!
So NOW the hypocrites are demanding permission to train a model on THEIR data.
What I get from this is frontier model capabilities are being stagnant.
The outputs belong to whoever purchased them. What are they complaining about?
I would say Antrophic and others illicitly extracted free internet content and put it behind a paywall, giving zero compensation to those that made their whole business possible in the first place. So smallest violin player busy here trying to make me care if it happens to them.
Thieves stealing from each other? No way. Guess on what anthropic trained their data.
I'm sorry, but I can't stop laughing at an AI company crying about theft of their IP.
A company which got rich on extracting the world's content is complaining that another company has extracted their work?!
LOL!
Get a grip, son.
Anthropic also trained on all human knowledge, so I'm okay with others distilling their models.
fucking lol. it is always funny when companies use opensource and other free for non commercial use - and plain old piracy - and then cry about the same practices.
Oh no, someone is profiting of the work of others?!
anyways...
Karma truly is a bitch
Nevermind government edicts & bans -- this seems like reason enough for them to require Know Their Customers, require ID, and shut of certain nations.
Failing to have done so seems to have allowed 25000 fake Chinese accounts to walk off with their product...
OFC I wouldn't trust the Chinese enough to ack their models the time of day, but Anthropic seems to have allowed far more ... yikes
People in glass houses shouldn't throw stones. Anthropic keeps throwing stones every few weeks.
Similar to improving an independent search engine by scraping Google search results and learning from it. Shady but legit
And Alibaba is releasing the full model weights open source under Apache 2.0, Anthropic… fuck that company.
How the hell does Anthropic continue to make such hypocritical complaints without deeply cringing?
It's becoming embarrassing to watch
Another day, another excuse as to why Fable 5 was pulled. Just waiting for Anthropic saying the Persona partnership was the fault of the Chinese.
> Meanwhile, on June 12, two days after Anthropic sent the letter, the Commerce Department imposed controversial restrictions on Anthropic's latest Mythos and Fable AI models because officials feared they could be deployed by military intelligence users in China and other countries of concern.
So that was the real reason for the Fable restriction? Because Anthropic wrote a letter to the US government saying that China was distilling Fable?
I am sorry, but companies doing biggest IP theft in history have no moral right to complain here.
"You're trying to kidnap what I've rightfully stolen!"
“Hey! Haven’t you heard that two wrongs don’t make a right?!”
- Entitled jerk that initially wronged people
You're teaching your parrot to say what our parrot is saying!
Good. I'm glad. Keep it up, China. Loving my cheap GLM and DeepSeek.
In an other news, a terrorist organization practicing torture at daily level just released a public denunciation of the evil forces they are fighting against, guided by their holy mission of making progress in social morality for all of us.
The Pot calling the Kettle black
"Hypocrisy, thy name is you"
willy wonka oh-go-on-dot-gif
Gosh, overusing accounts running up unplanned-for expenses?
Kinda reminds me of...overusage charges and inflated expenses clients have had to deal with because Anthropic, OpenAI, Grok, etc have been "illicitly extracting" everything they can grab from said websites, as fast as they can. In what amounts to a DDOS, frankly.
eh.. Anthropic wants open-weight models gone.
The pot calling the kettle black.
laughs in ironic
lol. good for the chinese. I hope their models get better than the closed american ones quick so we can stop using "controlled" models.
Being absolute ass to entire internet as you scour everything with no regard to common protocols - fine
Getting treated exactly same by competition - "we need rapid, coordinated action among industry players, policymakers and the global AI community."
Absolute scum. And the gall of going "oh buh it can be used for military, quick govt do something".
It's a bit disappointing when people see this as a schadenfreude moment because it's clearly not safe nor a good precedent for potentially dangerous AI models to trivially fall into the hands of malicious actors.
Oh, c'mon. If Alibaba wanted, it can have the entire Claude/Mythos source code and data by next week. All you need is enough bribe to a developer that has access to the repository. Humans are always the weakest link in anything.
How dare they! Only we should be illicitely extracting everything others done!
/Anthropic-probably
so what? anthropic stole this functionality from everyone else
Ok boomer
we now know what to use when Fable is too dangerous !
And that's coming from the intellectual property thieves. Laughable. Let the Chinese steal the models, they will only make it cheaper for everyone.
Well, of course they did. Are you kidding?
This article is absurd for an outlet who published an article that's meant to be news not editorial. Reuters was once a news wire and is still considered that. The first two paragraphs refer to "attack" and "strike" against Anthropic. This is sensational nonsense, not news. There was no strike, or attack. Block the accounts. Why is this news, and why are they pandering to the people who just banned the new model they burned at least $10 billion training? The closer you look at this AI stuff the more absurd it is. I assume the strat is to keep the bubble floating until post-2028, then drop the bomb on the Dem who wins. Just like with the covid inflation + economic rigging Trump did in 2016-2020.
Nobody cares. Grow up.
Why are they mad about this? Its not like they did not commit the biggest IP theft in modern history when training their models?
Oh, _now_ we care about IP, do we?
Cue Jeremy Clarkson's 'Oh no! Anyway...' GIF.
Suppose Anthropic trained only on data they paid to create, and not the internet or stolen textbooks.
It would still be extremely difficult to muster any sympathy for an organization whose MO is to go public not to honestly raise capital to fund growth and development, but rather to dishonestly leave someone else holding the bag, in some cases involuntarily as their retirement funds are passively invested.
And even supposing they were honest and didn't have an IPO, it would still be extraordinarily difficult to care about their misfortune, because "consolidating all thought-work into the hands of those few who can afford frontier models and datacenters and power plants" is also a special kind of misanthropy.
And even if that were not the case, they're filthy rich already, so who gives a shit if the Chinese companies prevent them from becoming quadrillionaires? :)
Ask claude it's name in chinese and it thinks its Qwen (opus) or Deepseek (sonnet). Anthropic are just as guilty as everyone else training AI, today, maybe more so. Every lab borrows from every other. It only takes a few hundred samples to figure out the pattern; look at glm-5.2 reasoning using the caveman tongue of gpt-5.5. Stopping this would require some draconian surveillance.
That's not how it works though. When you prepare the conversations for distillation, it's the most trivial and obvious first step to replace "Qwen" with "Claude" and vice versa. I doubt they'd simply forget to do it.
A model may misidentify itself due to the surrounding context. When a model is about to answer "I'm ...", what follows is a sorted list of probabilities for what the next token should be. In most models it's usually a list of popular model names: say, in the list, first comes Claude, then Qwen, then ChatGPT etc. Usually the "Claude" token would be the most probable token, say 70%. But if the surrounding context is in Chinese, the embeddings for "something to do with China" may nudge the combined embedding of the output token towards the "Qwen" embedding more ("China+Claude=Qwen" in the embedding space). Say, the probability for "Qwen" now becomes 60% instead of 10%.
If we also use high temperature for more "creativity", the token sampler now may choose "Qwen". It's not the most probable token still, but it was chosen because selecting the 2nd most probable token once in a while usually allows a model to explore unexpected "creative" paths, and 60% probability is good enough compared to 70%. It's basically a hallucination.
I once made an experiment: if I ban the word "Qwen" in the inference engine entirely, and ask Qwen "which model are you?", it happily starts announcing it's Claude 100% time, simply because "Claude" is the next most probable token after "Qwen" in this context.