The answer has nothing to do with rationing. When all software is designed to grab the most attention, serve the most adds, collect the most data, deliver the maximum amount of the cheapest content and generally be as exploitive of its users as possible, no amount of rationing or not rationing will effect it.
McMaster Carr, a company worth a little over a billion, sells metal, bolts, casters, and many other industrial bits and bobs. Their website is infinitely better than Amazons. Why? It has nothing to do with rationing and everything to do with trying to actual deliver a good customer experience.
Ironically, so many modern tech companies are arguably less incentivized to provide a good technology experience than non-tech companies, since the tech company's value/revenue primarily comes from directly exploiting their users for their data and/or attention. Improving the user experience frequently directly reduces their revenue. No amount of free LLM usage can allow a video serving company to both serve ad-free videos and make money from ads in their videos.
Though, maybe the irony only exists because we use the broad term tech company instead of the more generally accurate terms data harvesting company and attention consuming company.
I don't think any of this makes much sense. The problem that this article ignores is that people build different things with AI, not the same things but faster. They build more superficial things, more throwaway things, things that are tailor-made for chasing the buck, and it just doesn't feel genuine or interesting.
It was much the same with search. Google search allowed us to find things faster at first...but over time websites responded to that and started doing SEO crap. Now almost every search results in a billion websites that say the same thing, but shallowly. Yes you can still find some gems but was it worth it?
AI makes things faster, but now it means we can target superificial needs very quickly...and so a new breed of crap software will be built with it. As long as there's a dollar sign at the end of the road.
I don't understand this argument. Why do you think everyone is "chasing the buck"? That describes 0% of the work I've done over the last 6 months. There's no "bucks" in the music players, macro trackers, and source code review tools I've built.
AI makes things faster, and it pushes the envelope on what can be built. Like a telescope lets the human eye see further into space, an LLM helps track more ideas and structures within working memory. Yes, people will use this new potential to build slop. And other people will use it to build what has never been built before. The only question is which one we will prefer.
By definition an LLM is unable to build what has never been built before because you are going outside the realm of pretraining/RL. You might be able to combine components that already exist in a unique way, but even then you will be fighting against the mediocre making effect of LLMs
That understates what LLMs are capable of doing. They can and do build things outside of their training dataset. As long as you can get there by following existing concepts, they can do it.
Perhaps outside their training set in a strict sense of “this thing itself was not specifically in the training set” (although only perhaps - users often iteratively coax frontier models into achieving things just beyond their envelope, and those achievements could well be folded into the next model’s training set without you or I being aware of that). But even then, it is not directionally outside the training set: it steps outside not in the direction of the unknown, but in the direction of another part of its training set. An evocative metaphor would be dashing from cover to cover in the rain, because it lacks the umbrella or raincoat or simply the fortitude to tolerate getting wet that it would need to reach an uncovered destination.
I recall something I once heard about product marketing. The experiment was a product like jam, and they had 3 flavour options available in one case and 12 options available in another. It was a test on how hard it was to make a decision as well as how confident the person was they made the right decision later on.
One of Sartre's points on existentialism was that radical freedom leads to anxiety.
In some sense, AI is removing scarcity. It isn't clear to me that this is the same kind of thing as going from starvation to surplus. It is more like the jam case where we aren't increasing abundance in a total sense, we increasing the options within the existing surplus.
Those who haven't read Mythical Man Month are doomed to repeat it. I don't recall anywhere in the book where it states that the problems described in its essays don't apply to you if you're working with dozens of the most intelligent engineers in the world who can produce code in hours instead of days.
I was discussing this with someone else and used the analogy of nuclear energy. Although nuclear plants are highly complex and use forces that have only been understood for the last hundred years, at the end of the day it's still boiling water to create steam to spin a turbine, and nothing about nuclear innovation can engineer away the laws of thermodynamics.
Agents have and will continue to drastically change how we develop code, but it's my belief that drastically changing our software practices to speed up development just means we're going to hit engineering limits faster, rather than being able to overcome those limits entirely.
I was telling a friend something similar last week.
Unless, or perhaps until, we figure out how to use LLMs effectively at the organization level we're still going to be limited to the "slime mold" level of organizational intelligence.
Will that happen? Most likely, but will it take a generation or ten? I suspect it'll be far harder for society to adapt at that level. We've had Mythical Man Month for decades yet software managers continue repeating the same mistakes.
Because of that the idea of OAI and Anthropic just moving to in-house product development makes sense.
At least what I’m seeing at work is that AI tools have generated thousands of security reports after scanning all of the old code. So our whole team has been bogged down for months now removing eval(), updating ancient packages, updating docker images that haven’t been rebuilt in years, etc. It shows no improvement to the end user but since all of these vulnerabilities have suddenly become very visible we have to action them all right now.
People respond to incentives: Why do a hard thing when I can do an easy thing?
During the hype cycle, making "x for y" is easiest to get funded and exit.
Once the hype crashes and we're in the trough of disillusionment, then people are bearish. So founders either move to the next hype cycle or figure out a more durable business from the ashes
To be honest I don’t believe this is the case. Admittedly, I don’t have any experience with the best of the best models, only up to clause Opus. AI accelerates coding for myself a hell of a lot, but frankly the amount of time I spend coding complex programs feels about the same. I think there’s still a bottleneck, and it has been reiterated here before: judgement, testing, and understanding. If AI really accelerated coding, no amount of cultural inertia would be able to hold it back. It’s just not.
I was about to say something similar. The article’s premise is solid as long as we accept the assumption that AI is actually fundamentally changing the velocity and methodology of coding.
To me, so far it seems like it is and it isn’t simultaneously. Some things are going at insane paces. I see things like reverse engineering and emulation breaking barriers. But we don’t see things like the quality and complexity of highly architected systems getting much better, maybe even getting worse as management pushes velocity onto employees.
I think what this shows is that AI is great at tedium and brute forcing, but it’s not actually good at making substantially better software with less of that “rationed” expertise. At least, not yet.
All this talk about removing bottlenecks at all costs, and we haven't even considered the final boss of bottlenecks: the end users themselves. All the gains will be a waste if the end users have a choice. And as "The Goal" has taught us, we must remove the further-most bottlenecks first. Hence, we must remove all choice and power from end-users first.
"Build me a high speed railway. Make no mistakes."
Now that SaaS CRUD apps can be built in 10 minutes, we can maybe do something more worthwhile, no? So much wasted effort has gone into this field already, in my opinion.
... by improving model efficiency so AI tokens are "too cheap to meter" or alternately not that much more expensive to operate than conventional sorts of software?
It's a different view of scaling. I mean a lot of people have this absurd idea of "super intelligence" which is like the cover of this book
i mean what does "intelligence quotient 10,000" even mean? On the other hand it's very easy to imagine a world where we have systems like the best ones we have now but we can really afford to use them.
"So why isn't our software 10x better?"
The answer has nothing to do with rationing. When all software is designed to grab the most attention, serve the most adds, collect the most data, deliver the maximum amount of the cheapest content and generally be as exploitive of its users as possible, no amount of rationing or not rationing will effect it.
McMaster Carr, a company worth a little over a billion, sells metal, bolts, casters, and many other industrial bits and bobs. Their website is infinitely better than Amazons. Why? It has nothing to do with rationing and everything to do with trying to actual deliver a good customer experience.
Ironically, so many modern tech companies are arguably less incentivized to provide a good technology experience than non-tech companies, since the tech company's value/revenue primarily comes from directly exploiting their users for their data and/or attention. Improving the user experience frequently directly reduces their revenue. No amount of free LLM usage can allow a video serving company to both serve ad-free videos and make money from ads in their videos.
Though, maybe the irony only exists because we use the broad term tech company instead of the more generally accurate terms data harvesting company and attention consuming company.
I don't think any of this makes much sense. The problem that this article ignores is that people build different things with AI, not the same things but faster. They build more superficial things, more throwaway things, things that are tailor-made for chasing the buck, and it just doesn't feel genuine or interesting.
It was much the same with search. Google search allowed us to find things faster at first...but over time websites responded to that and started doing SEO crap. Now almost every search results in a billion websites that say the same thing, but shallowly. Yes you can still find some gems but was it worth it?
AI makes things faster, but now it means we can target superificial needs very quickly...and so a new breed of crap software will be built with it. As long as there's a dollar sign at the end of the road.
I don't understand this argument. Why do you think everyone is "chasing the buck"? That describes 0% of the work I've done over the last 6 months. There's no "bucks" in the music players, macro trackers, and source code review tools I've built.
AI makes things faster, and it pushes the envelope on what can be built. Like a telescope lets the human eye see further into space, an LLM helps track more ideas and structures within working memory. Yes, people will use this new potential to build slop. And other people will use it to build what has never been built before. The only question is which one we will prefer.
By definition an LLM is unable to build what has never been built before because you are going outside the realm of pretraining/RL. You might be able to combine components that already exist in a unique way, but even then you will be fighting against the mediocre making effect of LLMs
That understates what LLMs are capable of doing. They can and do build things outside of their training dataset. As long as you can get there by following existing concepts, they can do it.
Perhaps outside their training set in a strict sense of “this thing itself was not specifically in the training set” (although only perhaps - users often iteratively coax frontier models into achieving things just beyond their envelope, and those achievements could well be folded into the next model’s training set without you or I being aware of that). But even then, it is not directionally outside the training set: it steps outside not in the direction of the unknown, but in the direction of another part of its training set. An evocative metaphor would be dashing from cover to cover in the rain, because it lacks the umbrella or raincoat or simply the fortitude to tolerate getting wet that it would need to reach an uncovered destination.
That’s a hard position to defend in the face of navier stokes
I am not sure this is default a good thing.
I recall something I once heard about product marketing. The experiment was a product like jam, and they had 3 flavour options available in one case and 12 options available in another. It was a test on how hard it was to make a decision as well as how confident the person was they made the right decision later on.
One of Sartre's points on existentialism was that radical freedom leads to anxiety.
In some sense, AI is removing scarcity. It isn't clear to me that this is the same kind of thing as going from starvation to surplus. It is more like the jam case where we aren't increasing abundance in a total sense, we increasing the options within the existing surplus.
Those who haven't read Mythical Man Month are doomed to repeat it. I don't recall anywhere in the book where it states that the problems described in its essays don't apply to you if you're working with dozens of the most intelligent engineers in the world who can produce code in hours instead of days.
I was discussing this with someone else and used the analogy of nuclear energy. Although nuclear plants are highly complex and use forces that have only been understood for the last hundred years, at the end of the day it's still boiling water to create steam to spin a turbine, and nothing about nuclear innovation can engineer away the laws of thermodynamics.
Agents have and will continue to drastically change how we develop code, but it's my belief that drastically changing our software practices to speed up development just means we're going to hit engineering limits faster, rather than being able to overcome those limits entirely.
I was telling a friend something similar last week.
Unless, or perhaps until, we figure out how to use LLMs effectively at the organization level we're still going to be limited to the "slime mold" level of organizational intelligence.
Will that happen? Most likely, but will it take a generation or ten? I suspect it'll be far harder for society to adapt at that level. We've had Mythical Man Month for decades yet software managers continue repeating the same mistakes.
Because of that the idea of OAI and Anthropic just moving to in-house product development makes sense.
At least what I’m seeing at work is that AI tools have generated thousands of security reports after scanning all of the old code. So our whole team has been bogged down for months now removing eval(), updating ancient packages, updating docker images that haven’t been rebuilt in years, etc. It shows no improvement to the end user but since all of these vulnerabilities have suddenly become very visible we have to action them all right now.
People respond to incentives: Why do a hard thing when I can do an easy thing?
During the hype cycle, making "x for y" is easiest to get funded and exit.
Once the hype crashes and we're in the trough of disillusionment, then people are bearish. So founders either move to the next hype cycle or figure out a more durable business from the ashes
To be honest I don’t believe this is the case. Admittedly, I don’t have any experience with the best of the best models, only up to clause Opus. AI accelerates coding for myself a hell of a lot, but frankly the amount of time I spend coding complex programs feels about the same. I think there’s still a bottleneck, and it has been reiterated here before: judgement, testing, and understanding. If AI really accelerated coding, no amount of cultural inertia would be able to hold it back. It’s just not.
I was about to say something similar. The article’s premise is solid as long as we accept the assumption that AI is actually fundamentally changing the velocity and methodology of coding.
To me, so far it seems like it is and it isn’t simultaneously. Some things are going at insane paces. I see things like reverse engineering and emulation breaking barriers. But we don’t see things like the quality and complexity of highly architected systems getting much better, maybe even getting worse as management pushes velocity onto employees.
I think what this shows is that AI is great at tedium and brute forcing, but it’s not actually good at making substantially better software with less of that “rationed” expertise. At least, not yet.
All this talk about removing bottlenecks at all costs, and we haven't even considered the final boss of bottlenecks: the end users themselves. All the gains will be a waste if the end users have a choice. And as "The Goal" has taught us, we must remove the further-most bottlenecks first. Hence, we must remove all choice and power from end-users first.
Tell me I'm wrong.
hehe...Before Claude(BC) and After Claude(AC).
In Claude we trust.
"Build me a high speed railway. Make no mistakes."
Now that SaaS CRUD apps can be built in 10 minutes, we can maybe do something more worthwhile, no? So much wasted effort has gone into this field already, in my opinion.
... by improving model efficiency so AI tokens are "too cheap to meter" or alternately not that much more expensive to operate than conventional sorts of software?
It's a different view of scaling. I mean a lot of people have this absurd idea of "super intelligence" which is like the cover of this book
https://en.wikipedia.org/wiki/Supermind_(novel)
i mean what does "intelligence quotient 10,000" even mean? On the other hand it's very easy to imagine a world where we have systems like the best ones we have now but we can really afford to use them.
> i mean what does "intelligence quotient 10,000" even mean?
IQ tests measure, essentially, how fast you can do shallow pattern recognition.
And AI can do these kinds of tests easily.
There are time limits but to say they're primarily measuring speed sounds wrong.
Before we were spending 10:1 on features:quality . I worry that ratio will remain, but at 100x or 1000x the velocity.
There's no reason to think that companies are suddenly going to be spending 1:1 features : quality because both are faster / cheaper .
So expect the low quality to continue, and naturally get worse, because complexity is going up 1000x
The free holiday will last forever. Make your plans accordingly.
this reads like an ad.