Melanie Mitchell: Post-Training Like RLHF Makes Models No Longer Statistical Language Models

MelMitchell1 · x · 2026-09-27

Santa Fe Institute's Melanie Mitchell weighs into a terminology debate: people now use "LLM" loosely for any large AI system built on a pre-trained LM, which is not the sense meant in the Stochastic Parrots paper — that's her only point.

She adds that any post-training moving the model away from the general statistics of language (e.g., RLHF or RL for tool use) makes it no longer a statistical model of language in the sense of modeling the structure of human language.

Original post →

Melanie Mitchell: Post-Training Like RLHF Makes Models No Longer Statistical Language Models

More from Models