Freddie DeBoer has this thing where he asks: imagine your life without any form of indoor plumbing and imagine your life without LLMs. Which absence would have a bigger impact on your everyday experience and your quality of life? You can play this game with a bunch of everyday technologies: refrigeration, shipping containers, antibiotics, etc. Even something as mundane and limited as a home washing machine.
Sure, an LLM can spin up a power-point presentation for me in a few minutes without much effort on my part, but was that presentation really as consistently life-changing as having clean clothes without the attendant hours of physical drudgery?
There's another option here where LLMs are an end-state technology where the culture shifts focus to more social and spiritual meaning-making and maintains these kinds of information technologies only where they can directly serve our values, which are defined in human terms. This is the future that I am bullish on.
If we want to reduce this question to purely economic terms, we would see LLMs generating a substantial and long-run productivity boom, and, as far as I know, there's no evidence that this is happening, a half-decade into their commercial introduction. When thinking about the impact of technologies, too many people are distracted by these financialization schemes, which will eventually run their course.
I mean that’s the trillion-dollar question: how much value, defined broadly, is generated by this technology? Who will actually do it? And will they generate that value before the capex kills them?
Productivity is the thing to look at. Other economic metrics and personal-anecdote value propositions are a distraction. If these technologies don't drive an economy-wide productivity boom, once the hype recedes, there just won't be the enough income from other, more fundamental parts of the economy to sustain investment. LLMs aren't going away, but they will probably end up a being a drag on the balance sheets of the big tech companies -- they can't not offer LLM-enabled features in their products going forward, but LLMs can't attract enough additional revenues to make a profit.
Frankly, I think the hallucination problem makes this inevitable. You can't rely on LLMs to do anything that needs to be correct or needs to work reliably, and the LLM providers do not want to shoulder the accountability penalties that are normal in our economic and legal life. Also, a lot of work isn't necessarily structured like a prompt-and-output sequence.
I'd love it if you were the ISO-NE watcher. Such a great primer on this issue.
What if LLMs are not fake, but merely trivial?
Freddie DeBoer has this thing where he asks: imagine your life without any form of indoor plumbing and imagine your life without LLMs. Which absence would have a bigger impact on your everyday experience and your quality of life? You can play this game with a bunch of everyday technologies: refrigeration, shipping containers, antibiotics, etc. Even something as mundane and limited as a home washing machine.
Sure, an LLM can spin up a power-point presentation for me in a few minutes without much effort on my part, but was that presentation really as consistently life-changing as having clean clothes without the attendant hours of physical drudgery?
There's another option here where LLMs are an end-state technology where the culture shifts focus to more social and spiritual meaning-making and maintains these kinds of information technologies only where they can directly serve our values, which are defined in human terms. This is the future that I am bullish on.
If we want to reduce this question to purely economic terms, we would see LLMs generating a substantial and long-run productivity boom, and, as far as I know, there's no evidence that this is happening, a half-decade into their commercial introduction. When thinking about the impact of technologies, too many people are distracted by these financialization schemes, which will eventually run their course.
I mean that’s the trillion-dollar question: how much value, defined broadly, is generated by this technology? Who will actually do it? And will they generate that value before the capex kills them?
I unfortunately don’t have the answers
Productivity is the thing to look at. Other economic metrics and personal-anecdote value propositions are a distraction. If these technologies don't drive an economy-wide productivity boom, once the hype recedes, there just won't be the enough income from other, more fundamental parts of the economy to sustain investment. LLMs aren't going away, but they will probably end up a being a drag on the balance sheets of the big tech companies -- they can't not offer LLM-enabled features in their products going forward, but LLMs can't attract enough additional revenues to make a profit.
Frankly, I think the hallucination problem makes this inevitable. You can't rely on LLMs to do anything that needs to be correct or needs to work reliably, and the LLM providers do not want to shoulder the accountability penalties that are normal in our economic and legal life. Also, a lot of work isn't necessarily structured like a prompt-and-output sequence.