AI, used in small, local models, as an assistance tool, is actually somewhat helpful. AI is how Google Translate got so good a decade or so ago, for instance; and how assistive image recognition has become good enough that visually-impaired people can potentially access the web just as proficiently as sighted people. LLM-assisted spell check, grammar check, and autocomplete show a lot of promise. LLM-assisted code completion is already working decently well for common programming languages. There are potentially other halfway decent uses as well.
Basically, if you let computers do what they’re good at (objective, non-creative, repetitive, large-dataset tasks that don’t require reasoning or evaluation), they can make humans better at what they’re good at (creativity, pattern-matching, ideation, reasoning). And AI can help with that, even though they can’t get humans out of the loop.
But none of those things put dollar signs in VC’s eyes. None of those use cases get executives thinking, “hey, maybe we can fire people and save on the biggest single recurring expense any corporation puts on their balance sheet.” None of these make worried chip manufacturers breathe a sigh of relief that they can continue making the line go up after Moore’s Law finally kicks the bucket. None of those things make headlines in late-stage capitalism. Elon Musk can’t use any of those things as smokescreens to distract from his mismanagement of the (formerly) most consequential social media brand in history. None of that gives former crypto bros that same flutter of superiority.
So the hype gets pumped up to insane levels, which makes the valuations inflate, which makes them suck up more data heedless of intellectual property, which makes them build more power-hungry data centers, which means they have to generate more hype (based on capabilities the technology emphatically does not have and probably never will) to justify all of it.
Like with crypto. Blockchain showed some promise in extremely niche, low-trust environments; but that wasn’t sexy, or something that anyone could sell.
Once the AI bubble finally breaks, we might actually get some useful tools out of it. Maybe. But you can’t sell that.
AI, used in small, local models, as an assistance tool, is actually somewhat helpful. AI is how Google Translate got so good a decade or so ago, for instance; and how assistive image recognition has become good enough that visually-impaired people can potentially access the web just as proficiently as sighted people. LLM-assisted spell check, grammar check, and autocomplete show a lot of promise. LLM-assisted code completion is already working decently well for common programming languages. There are potentially other halfway decent uses as well.
Basically, if you let computers do what they’re good at (objective, non-creative, repetitive, large-dataset tasks that don’t require reasoning or evaluation), they can make humans better at what they’re good at (creativity, pattern-matching, ideation, reasoning). And AI can help with that, even though they can’t get humans out of the loop.
But none of those things put dollar signs in VC’s eyes. None of those use cases get executives thinking, “hey, maybe we can fire people and save on the biggest single recurring expense any corporation puts on their balance sheet.” None of these make worried chip manufacturers breathe a sigh of relief that they can continue making the line go up after Moore’s Law finally kicks the bucket. None of those things make headlines in late-stage capitalism. Elon Musk can’t use any of those things as smokescreens to distract from his mismanagement of the (formerly) most consequential social media brand in history. None of that gives former crypto bros that same flutter of superiority.
So the hype gets pumped up to insane levels, which makes the valuations inflate, which makes them suck up more data heedless of intellectual property, which makes them build more power-hungry data centers, which means they have to generate more hype (based on capabilities the technology emphatically does not have and probably never will) to justify all of it.
Like with crypto. Blockchain showed some promise in extremely niche, low-trust environments; but that wasn’t sexy, or something that anyone could sell.
Once the AI bubble finally breaks, we might actually get some useful tools out of it. Maybe. But you can’t sell that.