DeepMind vet Graepel: ten years after AlphaGo, LLMs still don't reason
nordicinst · x · 2026-10-02
DeepMind researcher and AlphaGo team member Thore Graepel published an MIT Technology Review opinion piece marking ten years since AlphaGo beat Lee Sedol, questioning whether LLMs truly reason.
- What made Move 37 work: AlphaGo didn't win by luck — it "chain-smoked its own future," searching through auditable reasoning paths via self-play, enabling it to invent moves no human had imagined.
- Contrast with LLMs: today's models chase "good vibes" rather than verifiable reasoning; their apparent reasoning is post-hoc rationalization, not the auditable path AlphaGo relied on.
- His prescription: true reasoning needs an auditable path, not a victory lap — "audit, reason, repeat," which he also frames as a direction for AI (and Europe's AI future).
Related event: AlphaGo Veteran Questions LLM Reasoning, Sparking Researcher Debate(6 posts)→
More from AGI Musings
- Top researchers are quietly stopping paper submissions, observer notes — avt_im · 2026-11-17
- "Abundant intelligence should terrify businesses built on keeping it scarce" — _AustinCalvert_ · 2026-11-17
- 130-page paper introduces "bottleneck pressure": work AI can't do gains value as the rest speeds up — soumitrashukla9 · 2026-11-17
- AI conference authors and reviewers complain unprecedentedly about AI slop — khademinori · 2026-11-17
- Sam Altman: AI already speeds up OpenAI's research, but a self-improving button? 'Definitely not' — ChrisGPT · 2026-11-17
- "Financiers can remain unconvinced longer than you can remain a techno visionary" — shakoistsLog · 2026-11-17