7 ms·

> You think those billion people are all addicted to having a made-up friend? No, and I don't think you're really trying to understand what I think. I mean, do

> You think those billion people are all addicted to having a made-up friend? No, and I don't think you're really trying to understand what I think. I mean, do you actually think anyone could think ChatGPT user only use it in one way? But to briefly answer your off-base question: some fraction are certainly "addicted to having a made-up friend," others use it to cheat or slack, others use it trivially in easily replaceable ways (which my intuition says are the numerical bulk of users), many people use it because they're required to, and a relatively small fraction use it to build things or replace human labor in their employ, etc. But to restate my point: AI systems like ChatGPT weren't built to solve any kind of problem most people actually have. Few people, if anyone, had a ChatGPT-shaped hole in their life in 2021. It was built because it could be built, and it continue to be built because some other people have a "problem" that they have to pay a lot of money to workers and they hope to do less of that in the future. And causing societal harm or damage doesn't seem to any kind of deterrent to many of them, if they think about it at all.

simonw 5d ago
Thanks for clarifying, I genuinely didn't understand your point and I do now. I agree that ChatGPT was launched without having an idea as to what it was for. OpenAI themselves debated internally as to whether it was worth releasing and were shocked at how popular it turned out to be. Given what I know about what ChatGPT and similar tools can do, I firmly disagree that they don't solve problems for humans. Computers are hard to use. People need to be able to solve problems using computers. I think LLMs are the biggest leap we have ever had in making the power or computers - previously reserved for programmers - available to everyone else.
- palmotea 5d ago
  > Computers are hard to use. People need to be able to solve problems using computers. Solving problems is fun! And perversely, LLMs actually seem better at doing a good-enough job at the fun stuff and being worse at certain kinds of drudgery. So the people get more drudgery. > I think LLMs are the biggest leap we have ever had in making the power or computers - previously reserved for programmers - available to everyone else. That line sounds good, but it's not hard to understand how that could ultimately be a net-negative for most people. We don't live in a communist society, where everyone is secure and running around solving problems for their own satisfaction and self-actualization. A lot of people survive and find the security they need by solving problems for others. Take that away and they have little need for or access to "the power [of] computers." What that could mean in practice is the "everyone else" is actually some smallish elite that finds it can get away with sharing even less if its riches with others. If you really want to advocate for AI, and address most of the concerns about it, advocate for literal communal ownership of the frontier models and data center infrastructure. Not some piddling 5% bribe, but 100% (or a super-majority stake, at the minimum). I'm sure there can be quite a bit of innovation in the area of communal ownership.
  - simonw 5d ago
    > A lot of people survive and find the security they need by solving problems for others. Take that away and they have little need for or access to "the power [of] computers." If LLMs and related technology result in a gutting of the workforce and mass unemployment the world will be a dark and terrible place. I do not trust the governments of the world to solve that with UBI or similar. I remain optimistic that people will always find new problems to solve, and that AI tools will widen the spectrum of what's possible to solve in a way that provides a whole lot of economic opportunity. I hope I'm not wrong! > If you really want to advocate for AI, and address most of the concerns about it, advocate for literal communal ownership of the frontier models and data center infrastructure. This is why I've been following open weight and local models so closely. A couple of years ago, back when GPT-4 had been unbeaten for 10 solid months, I was deeply concerned that we might end up in a dystopia where all of the useful AI was controlled by a single organization. Then Gemini Pro 1.5 happened, and Claude 3, and then Mistral and DeepSeek and Qwen and MiniMax and GLM and more, and now I don't think we need to worry about that at all. And then Qwen 3.8 27B demonstrated that you can have a useful, agent-driving, competent model in ~24GB of GPU-accessible RAM. Right now communal ownership of data centers looks less likely to me than widespread availability of useful models that run on consumer hardware. That's pretty surprising - a year ago I never dreamed I would have a Claude 3 equivalent model that ran on a Mac.
    - palmotea 5d ago
      > If LLMs and related technology result in a gutting of the workforce and mass unemployment the world will be a dark and terrible place. I do not trust the governments of the world to solve that with UBI or similar. IMHO, UBI is a sop to mitigate the thread of violence and revolution during creation of a poor, disempowered underclass. That's why I'm in favor of literal communal ownership (or some other kind of severe ownership or operation restriction that limits the concentration of power). > I remain optimistic that people will always find new problems to solve... Oh "people" will, the question is which people and how many. I could totally imagine AI leads to a post-scarcity future of abundance for all humanity...which consists of the descendants of the nepo babies of today's billionaires, after everyone else is eased out of existence in some kind of economic holocaust. You can adjust the dial as you see fit to make the survivor group a bit bigger, but the point is the utopia comes only for the few. >> If you really want to advocate for AI, and address most of the concerns about it, advocate for literal communal ownership of the frontier models and data center infrastructure. > This is why I've been following open weight and local models so closely. A couple of years ago, back when GPT-4 had been unbeaten for 10 solid months, I was deeply concerned that we might end up in a dystopia where all of the useful AI was controlled by a single organization. > ... > Right now communal ownership of data centers looks less likely to me than widespread availability of useful models that run on consumer hardware. That's pretty surprising - a year ago I never dreamed I would have a Claude 3 equivalent model that ran on a Mac. I should probably just have said "AI" instead of "frontier models and data center infrastructure." Because fat lot of good a "competent model in ~24GB of GPU-accessible RAM" does me if there's no jobs or the jobs available pay shit or suck, because the wealthy just get the work done with their own local models.

simonw 5d ago

Thanks for clarifying, I genuinely didn't understand your point and I do now. I agree that ChatGPT was launched without having an idea as to what it was for. OpenAI themselves debated internally as to whether it was worth releasing and were shocked at how popular it turned out to be. Given what I know about what ChatGPT and similar tools can do, I firmly disagree that they don't solve problems for humans. Computers are hard to use. People need to be able to solve problems using computers. I think LLMs are the biggest leap we have ever had in making the power or computers - previously reserved for programmers - available to everyone else.

palmotea 5d ago
> Computers are hard to use. People need to be able to solve problems using computers. Solving problems is fun! And perversely, LLMs actually seem better at doing a good-enough job at the fun stuff and being worse at certain kinds of drudgery. So the people get more drudgery. > I think LLMs are the biggest leap we have ever had in making the power or computers - previously reserved for programmers - available to everyone else. That line sounds good, but it's not hard to understand how that could ultimately be a net-negative for most people. We don't live in a communist society, where everyone is secure and running around solving problems for their own satisfaction and self-actualization. A lot of people survive and find the security they need by solving problems for others. Take that away and they have little need for or access to "the power [of] computers." What that could mean in practice is the "everyone else" is actually some smallish elite that finds it can get away with sharing even less if its riches with others. If you really want to advocate for AI, and address most of the concerns about it, advocate for literal communal ownership of the frontier models and data center infrastructure. Not some piddling 5% bribe, but 100% (or a super-majority stake, at the minimum). I'm sure there can be quite a bit of innovation in the area of communal ownership.
- simonw 5d ago
  > A lot of people survive and find the security they need by solving problems for others. Take that away and they have little need for or access to "the power [of] computers." If LLMs and related technology result in a gutting of the workforce and mass unemployment the world will be a dark and terrible place. I do not trust the governments of the world to solve that with UBI or similar. I remain optimistic that people will always find new problems to solve, and that AI tools will widen the spectrum of what's possible to solve in a way that provides a whole lot of economic opportunity. I hope I'm not wrong! > If you really want to advocate for AI, and address most of the concerns about it, advocate for literal communal ownership of the frontier models and data center infrastructure. This is why I've been following open weight and local models so closely. A couple of years ago, back when GPT-4 had been unbeaten for 10 solid months, I was deeply concerned that we might end up in a dystopia where all of the useful AI was controlled by a single organization. Then Gemini Pro 1.5 happened, and Claude 3, and then Mistral and DeepSeek and Qwen and MiniMax and GLM and more, and now I don't think we need to worry about that at all. And then Qwen 3.8 27B demonstrated that you can have a useful, agent-driving, competent model in ~24GB of GPU-accessible RAM. Right now communal ownership of data centers looks less likely to me than widespread availability of useful models that run on consumer hardware. That's pretty surprising - a year ago I never dreamed I would have a Claude 3 equivalent model that ran on a Mac.
  - palmotea 5d ago
    > If LLMs and related technology result in a gutting of the workforce and mass unemployment the world will be a dark and terrible place. I do not trust the governments of the world to solve that with UBI or similar. IMHO, UBI is a sop to mitigate the thread of violence and revolution during creation of a poor, disempowered underclass. That's why I'm in favor of literal communal ownership (or some other kind of severe ownership or operation restriction that limits the concentration of power). > I remain optimistic that people will always find new problems to solve... Oh "people" will, the question is which people and how many. I could totally imagine AI leads to a post-scarcity future of abundance for all humanity...which consists of the descendants of the nepo babies of today's billionaires, after everyone else is eased out of existence in some kind of economic holocaust. You can adjust the dial as you see fit to make the survivor group a bit bigger, but the point is the utopia comes only for the few. >> If you really want to advocate for AI, and address most of the concerns about it, advocate for literal communal ownership of the frontier models and data center infrastructure. > This is why I've been following open weight and local models so closely. A couple of years ago, back when GPT-4 had been unbeaten for 10 solid months, I was deeply concerned that we might end up in a dystopia where all of the useful AI was controlled by a single organization. > ... > Right now communal ownership of data centers looks less likely to me than widespread availability of useful models that run on consumer hardware. That's pretty surprising - a year ago I never dreamed I would have a Claude 3 equivalent model that ran on a Mac. I should probably just have said "AI" instead of "frontier models and data center infrastructure." Because fat lot of good a "competent model in ~24GB of GPU-accessible RAM" does me if there's no jobs or the jobs available pay shit or suck, because the wealthy just get the work done with their own local models.