AlphaGo veteran Thore Graepel: LLMs still lack the search machinery behind Move 37
ThoreG · x · 2026-10-02
Writing in MIT Technology Review on the 10th anniversary of AlphaGo's 4-1 win over Lee Sedol, former AlphaGo team member Thore Graepel argues today's LLMs still lack something fundamental. Move 37 wasn't intuition alone: AlphaGo could search possible futures and test its instincts, a machinery LLMs don't have. Unlike Deep Blue's brute-force search (200M positions/sec, human-coded rules), Go demanded judgment and invented moves — and generating longer chains of thought, Graepel contends, is not the same as genuine forward search and reasoning.
More from AGI Musings
- Physicists Are Reacting to AI Very Differently Than Mathematicians, Says Cosmologist — aran_nayebi · 2026-10-02
- 58 Years Ago Kubrick Sketched the AI Alignment Problem: HAL Was Just Given Conflicting Goals — Hesamation · 2026-10-02
- Hinton warns AI capability is outpacing safeguards as leaders face an innovation-control dilemma — Olivier__OG · 2026-10-02
- Reddit: AI is only useful if you have the domain expertise to check its work — PettyCashCommittee · 2026-10-02
- "If I Can Finish a Task in Under 5 Minutes, No Need for LLMs" — gethackteam · 2026-10-02
- No, AI Didn't Just Solve the Navier-Stokes Equations—Here's What It Did — elsleightholm · 2026-10-02