Researchers Debate Whether "Stochastic Parrot" Still Applies to Modern LLMs
On September 27, Redwood Research researcher Blanche Minerva and cognitive scientist Melanie Mitchell debated on social media over whether LLMs are merely "stochastic parrots" or next-token predictors.
Minerva proposed a five-step pipeline to rebut the claim: take a transformer, pretrain it on massive text, teach it special tokens for tool use, teach it to run programs via the command line, let it do long reasoning with CoT, and finally train it with RL — then she asked whether such a system still counts as a "stochastic parrot."
Confirmed
- Minerva argued that even pure LLMs no longer fit the "stochastic parrot" model, and several original authors have publicly stated the framework no longer applies; she noted the claim may have held around 2021 but is hard to justify today.
- Mitchell responded that she and Emily Bender had written a piece "in the opposite direction" explaining what "stochastic parrot" refers to when applied to LLMs; but she also said she has seen no one apply this critique to RL post-trained models, calling the "stochastic parrot" criticism a straw man in current discussions.
Why it matters
"Stochastic parrots" is one of the most widely circulated critical frameworks in the LLM field. This discussion shows that both its originators and critics agree the claim no longer fully applies to modern models trained with tool use, long chains of reasoning, and RL. Behind the terminology dispute lies a genuine disagreement over the substance of LLM capabilities.
2026-09-27 ~ 2026-09-27 · 6 related posts
Primary sources
- Blanche Minerva's five-step pipeline shows why "just next-token prediction" no longer fits LLMs — BlancheMinerva ·
- Melanie Mitchell: 'Stochastic parrots' is a strawman against RL-post-trained models — MelMitchell1 ·
- Researcher pushes back on 'stochastic parrot' framing: even pure LLMs no longer fit it — BlancheMinerva ·
- [source] Researcher pushes back on 'stochastic parrot' framing: even pure LLMs no longer fit it — BlancheMinerva · 2026-09-27
- Mitchell points to his and Emily Bender's essays on how "stochastic parrot" refers to LLMs — mmitchell_ai · 2026-09-27
- [source] Melanie Mitchell: 'Stochastic parrots' is a strawman against RL-post-trained models — MelMitchell1 · 2026-09-27
- [source] Blanche Minerva's five-step pipeline shows why "just next-token prediction" no longer fits LLMs — BlancheMinerva · 2026-09-27
- Researcher asks: a transformer with a command line — still a 'stochastic parrot'? — BlancheMinerva · 2026-09-27
- Dev challenges "stochastic parrot" critics: a transformer with a shell — how many lines until it's one? — BlancheMinerva · 2026-09-27