Have you tried this with AI with a topic you are well versed in? I am very optimistic AIs will help people feel they have some sort of understanding but far less optimistic they’ll actually help build it.
They’ll try to coerce any of your (or the model’s) mistakes into the correct answer out of politeness, and when you make a mistake it has zero model of the model you used to produce such a mistake, so it has no idea where to intervene in your knowledge except at which words hit the keyboard.
Idk, I am extremely extremely unimpressed any time I try to quiz it on topics I know. Except for coding, which makes sense since that’s the exercise of pushing as much semantic information into syntactic information as possible.
They’ll try to coerce any of your (or the model’s) mistakes into the correct answer out of politeness, and when you make a mistake it has zero model of the model you used to produce such a mistake, so it has no idea where to intervene in your knowledge except at which words hit the keyboard.
Idk, I am extremely extremely unimpressed any time I try to quiz it on topics I know. Except for coding, which makes sense since that’s the exercise of pushing as much semantic information into syntactic information as possible.