Comment by it_citizen

From what I understand of large language models, it won't play the best move it can, but the most likely to be played based on its training set.

If it was trained on chess.com, ChatGPT average level would probably be, at best, as good as the most average game it saw.

Correct?