Melanie Mitchell vs DeepMind Co-founder: Do Today's Models Still Count as 'Stochastic Parrots'?
_aidan_clark_ · x · 2026-09-27
Melanie Mitchell and DeepMind co-founder Aidan Clark debate on X: Mitchell notes the Stochastic Parrots paper used the original definition — a language model as a statistical model of language, not of everything post-training adds. Clark counters that everyone would have kept calling today's systems LLMs, and nobody said the next step would be fundamentally different. The exchange probes whether the stochastic parrot critique still applies.
More from AGI Musings
- Floridi et al. prove AI can't have guaranteed correctness and open-ended generality at once — rvp · 2026-09-27
- Dev Claim: Open Models on Consumer Hardware Can Handle 99.9% of AI Use Cases — ostrisai · 2026-09-27
- Anthropic's AI biolab deploys ~950 agents for 21 hours, finds CRISPR-like DNA in viruses — rvp · 2026-09-27
- Prof. Dietterich: Scaling Wasn't All You Needed — Data, RLVR and Harnesses Mattered — tdietterich · 2026-09-27
- AI access breaks creators' feedback loops, argues OpenAPS founder — scottleibrand · 2026-09-27
- GenAI Makes Producing 10x Easier, but Good Stuff Only 10% Easier — charles_irl · 2026-09-27