Melanie Mitchell: 'Stochastic parrots' is a strawman against RL-post-trained models

MelMitchell1 · x · 2026-09-27

Cognitive scientist Melanie Mitchell argues that claims calling today's RL-post-trained models with extensive harnesses "stochastic parrots" are a strawman—she sees no one seriously making that charge about models that solve hard math problems. Blanche Minerva pushed back that several original authors loudly maintain the claim, and that while it may have been largely true in 2021, even "pure" LLMs are no longer meaningfully captured by that framing.

Related event: Researchers Debate Whether "Stochastic Parrot" Still Describes LLMs(6 posts)→

Original post →

More from AGI Musings

AGI Musings channel →