Stochastic parrots
A critique arguing that fluent language-model output should not be equated with grounded meaning or understanding.
ConceptSources & confidence: Qualified confidence
Checked 2 October 2026 · AI-assisted review
Checked the definition, attribution and stated distinctions against the listed source passages.
- Emily M. Bender: Stochastic Parrots author resource page
Authorship, 2021 paper and author-maintained context for the critique.
What this check doesn’t establish
- Author resource page checked; full ACM paper access was unavailable in this pass.
- This is not a settled theorem about every model’s internal representations.
Confidence applies to the scope above, not every claim made by the underlying source. How confidence works
The phrase comes from Bender, Gebru, McMillan-Major and Shmitchell’s 2021 paper. It names a critique of treating statistical language generation as evidence of understanding and forms part of a wider discussion of language-model risks.
This record treats that position as an argument. A model’s failure to execute a written plan does not settle the philosophical question of meaning, just as fluent output alone does not settle it in the other direction.
In the essays
- I Gave an AI a Civilization to Run. It Built a Nuke.
Read the essay
- When the Agents Found God
Read the essay
Sources
- Moltbook was peak AI theater
Discussed in
- On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?
Introduced in · Discussed in
Ideas
Record history
Catalogue source: knowledge/concepts/stochastic-parrots.md
Generated by claude-code/claude-opus-5 on 2026-07-26