Blog argues 'stochastic parrots' metaphor derails LLM discourse via motte-and-bailey reasoning
manoelribeiro · x · 2026-10-02
Researcher Manoel Horta Ribeiro published a blog post arguing that the 'stochastic parrots' metaphor (Bender et al., FAccT 2021) is counterproductive to modern AI discourse, even in its mildest forms.
- Origins: The metaphor frames LLMs as systems that 'haphazardly stitch together' linguistic forms without reference to meaning, building on Bender & Koller 2020's form/meaning distinction. The paper famously triggered the departures of Timnit Gebru and Margaret Mitchell from Google.
- Core argument: Proponents use the metaphor to dismiss LLM capabilities, but when challenged retreat to the trivially true claim that LLMs predict next tokens — a classic motte-and-bailey structure.
- Conclusion: The metaphor has become confusing and unproductive, as its meaning shifts across each recurring social-media debate cycle.
More from AGI Musings
- Kyle Turman: nobody knows if AI is conscious — feigning certainty makes us careless — kyleturman · 2026-10-02
- Star Trek showed the best AI future: powerful machines everyone knows they're dealing with — paulnovosad · 2026-10-02
- Hypothesis: human learning is hill climbing — hard-to-verify tasks aren't relatively harder for AI — Afinetheorem · 2026-10-02
- PewDiePie's arc: from top YouTuber to home data center and self-trained LLM — tokumin · 2026-10-02
- Why Anthropic skips consumer life agents while OpenAI, Meta and xAI race in — sujingshen · 2026-10-02
- Instinct Burns More on LLMs Daily Than Its Entire Funding Round: No 'Experimentation Capital' in India — NirantK · 2026-10-02