Comment by joelthelion

> is just the next most likely token, taking into account temperature and what not.

This doesn't mean anything. All LLM output is like that.

That said, I agree that LLMs are terrible at grading stuff, except perhaps if you give them a very detailed evaluation grid.