I'd be interested to hear an opinion from Anthropic on this; there are two angles: why would they make content-related decisions going beyond any lawful use, and second, wouldn't this specific issue be antithetical to how LLMs work? Training frontier models was effectively the single largest instance of piracy in human history. I'm not passing judgement on whether that is good or bad, just calling it what it is, and saying that re-creations are objectively less controversial than that.
Is there any calculation/projection/cost-benefit done on the effects of these measures? I did not renew my Google AI sub because Gemini started to reject transcribing song lyrics due to IP. This seems like a similar thing
maybe this is what will get people to use grok or chinese models. At some point they're too restrictive. Claude refused to help me scrape a public website they themselves have gotten their info from..
Open-weight models are close. Claude is very good, but the others aren't far behind. The [cybersecurity] [cyber] alerts are annoying enough we looked for other options. Kimi and GLM have been very capable working with Ghidra, but I haven't tried REA yet
I honestly wonder if the "slopcomp"/"vibeport" (AI-heavy fan recomps/ports of old games) boom was getting into a problem for them at this point. Many of these things are also glorified single-game emulators, even having components from established emulators for things such as graphics rendering.
While I agree that lowlife opportunists using generative AI to front-run game creators is scummy, having the AI playing cop is not the right thing to do. There are any number of posts on HN describing misfires of these rules blocking legitimate user requests.
The AI bros want to liken AI to a compiler (*cough*BS*cough*); well, I've never had a compiler try to decline to compile code on moral grounds.
I oversee and contribute to compilers / toolchains, and I have sometimes had worries about what 'my' compiler could contribute to. I am of course not responsible, but it's not a nice feeling to know a path towards some unpleasant end had steps that involved you.
It reminds me of that version of Linux that was used for automated gun targeting and firing, perhaps in the early 2000s. Authors gave away their code as open source, and then that open source was used for what many felt was an evil purpose. A path to something many disliked or opposed now included them and their actions, despite their good intentions.
Sure, this rule is for a good cause (protecting small games developers) but it is a rule about protecting commercial interests of a group. One can imagine how it might be extended to protect larger, less worthy commercial interests. Imagine a compiler that declined to compile Linux or *BSD because they replicate the functionality of Solaris / HP-UX / AIX / Windows.
I'd be interested to hear an opinion from Anthropic on this; there are two angles: why would they make content-related decisions going beyond any lawful use, and second, wouldn't this specific issue be antithetical to how LLMs work? Training frontier models was effectively the single largest instance of piracy in human history. I'm not passing judgement on whether that is good or bad, just calling it what it is, and saying that re-creations are objectively less controversial than that.
Lawful use isn't a clear line.
Is there any calculation/projection/cost-benefit done on the effects of these measures? I did not renew my Google AI sub because Gemini started to reject transcribing song lyrics due to IP. This seems like a similar thing
maybe this is what will get people to use grok or chinese models. At some point they're too restrictive. Claude refused to help me scrape a public website they themselves have gotten their info from..
The entire issue with this is the dmca and not being able to share what is RE’d from the US.
So much more would be possible were this the case.
That being said… I’m not sure it should be illegal to share this re’d code.
And at the very least … just take it and rewrite what it does in rust lol.
Might not be clean room but hell, if it works.
I’ve gotten good mileage out of binary patching in trampolines lol
And adding in kernel shims.
The only part that’s missing that would probably help is tossing in more storage and moving logs off of root lol.
Open-weight models are close. Claude is very good, but the others aren't far behind. The [cybersecurity] [cyber] alerts are annoying enough we looked for other options. Kimi and GLM have been very capable working with Ghidra, but I haven't tried REA yet
And this, gives their competitors the edge in this regard, it doesn't stop it from happening.
I honestly wonder if the "slopcomp"/"vibeport" (AI-heavy fan recomps/ports of old games) boom was getting into a problem for them at this point. Many of these things are also glorified single-game emulators, even having components from established emulators for things such as graphics rendering.
Last time around on that front: https://www.pcgamer.com/gaming-industry/pc-ports-of-old-cons...
And remember the Samsung smart TV thing from days ago?
While I agree that lowlife opportunists using generative AI to front-run game creators is scummy, having the AI playing cop is not the right thing to do. There are any number of posts on HN describing misfires of these rules blocking legitimate user requests.
The AI bros want to liken AI to a compiler (*cough*BS*cough*); well, I've never had a compiler try to decline to compile code on moral grounds.
I oversee and contribute to compilers / toolchains, and I have sometimes had worries about what 'my' compiler could contribute to. I am of course not responsible, but it's not a nice feeling to know a path towards some unpleasant end had steps that involved you.
It reminds me of that version of Linux that was used for automated gun targeting and firing, perhaps in the early 2000s. Authors gave away their code as open source, and then that open source was used for what many felt was an evil purpose. A path to something many disliked or opposed now included them and their actions, despite their good intentions.
Sure, this rule is for a good cause (protecting small games developers) but it is a rule about protecting commercial interests of a group. One can imagine how it might be extended to protect larger, less worthy commercial interests. Imagine a compiler that declined to compile Linux or *BSD because they replicate the functionality of Solaris / HP-UX / AIX / Windows.
It would if the people who owned the compiler had enough market power. Thank gnu for keeping us out of that world.
Next, they’ll order it to stop remaking software.
right into the arms of open models and pi.dev
[dupe] https://news.ycombinator.com/item?id=50304358
another reason not to pay for those products