3 ms·
> 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 corr
by energy123 1y 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.