What I didn't predict is so many AI boosters being so miffed at regular folks pointing out the obvious: a half-working system may be tremendously exciting to AI enthusiasts, but it's not that useful to society.
Recently we've had all these people complaining of moving goalposts, when in fact most people never had goalposts for AI in the first place, and certainly never had goalposts that remotely resembled those that AI experts use to evaluate their own progress.
My mental model seems to clash with that of the boosters. They appear to have little doubt that reliability will emerge from scaling the models up. I think reliability is the lion's share of the problem and probably won't be solved by current architectures. It will be interesting to see who is right.
Recently we've had all these people complaining of moving goalposts, when in fact most people never had goalposts for AI in the first place, and certainly never had goalposts that remotely resembled those that AI experts use to evaluate their own progress.
My mental model seems to clash with that of the boosters. They appear to have little doubt that reliability will emerge from scaling the models up. I think reliability is the lion's share of the problem and probably won't be solved by current architectures. It will be interesting to see who is right.