Comment by energy123

6 months ago

> current-style LLMs, being inherently predictors of what a human would say

That's no longer what LLMs are. LLMs are now predictors of the tokens that are correlated with the correct answer to math and programming puzzles.