|
I think the problem here is that most of his claims are obvious, uninteresting, and largely agreed with even by the biggest AI hype people, like that AI hallucinates, or that they don't perfectly follow guardrails in the system prompt, or that they can be prompt injected, or that Open AI's financials look bad. But then on the other hand, he completely ignores all of the developments in the field scaffolding around these systems in order to resolve these problems. All of the changes and developments in how these models are trained, all of the things they've actually been able to achieve and do, and basically all of the positive use cases and things that balance out his criticisms. Since he doesn't really talk about any of that, of course he doesn't make false claims about it, he just ignores it, implicitly creating a false picture. And then it is this false picture that he uses to justify his grandiose claims about how everyone should have listened to him about how to do AI and these systems are inevitably going to turn out to be useless and the whole industry is going to collapse and fully disappear and society is going to be ruined and so on. So, of course, it looks like, on the one hand, all of his specific claims about AI are perfectly correct, and on the other hand, that all of his grinder claims about what that implies or means about the industry you have turned out to be wrong, and that he spends much more time on the latter than the former. I think it is really crucial to emphasize that even though most of the individual claims he makes are correct, he spends much more time on the prognostications that are fundamentally not correct, or at least are very speculative right now. I think that's an indication of something gone very wrong with someone's epistemic and incentive situation. |