I would like to ask the following question on MESE:
Some ChatGPT answers have been posted on (and removed from) MathOverflow, and there was a resulting MMO discussion.
In that discussion, @darijgrinberg said:
When I tell it about its errors, it "recognizes" them, sometimes elaborating on a counterexample I provide, but then it repeats the same errors again. I have a hunch that it treats lexically similar "words" as semantically similar, which dooms it to uselessness in mathematics.
I suspect I'm not alone in recognising common student behaviour when I read that description. Indeed, I know I'm not alone. @YemonChoi said:
@darijgrinberg " I have a hunch that it treats lexically similar "words" as semantically similar" - by complete coincidence reading this made me think of marking HW and some referee work I should be doing.
Given this (superficial but striking) similarity in behaviour, might the study of how AI "learns" mathematics shed any light on human learning?
This is very speculative, is based on little to no research on my part, and probably would not be something that could easily be put into immediate effect. I don't know the norms of what sorts of questions are appropriate on MESE. Could this one do well here as phrased? If not, are there any changes that could make it better, or is it simply entirely inappropriate?
EDIT: I have posted the question at Pedagogical insights to be gleaned from AI attempts to ‘learn’ mathematics (a few days ago, sorry—I forgot to edit earlier).
