Comment by mitxela

Some concrete facts about LLMs are explained by their next token predictor nature. Every time it says "wait, that's wrong." instead of generating the correct thing the first time.