LeCun says autoregressive CoT is not search; giffmana pushes back
giffmana · x · 2026-10-10
François Fleuret, citing LeCun, argues that auto-regressive prediction—stochastic or not—does not constitute search, since search requires examining and evaluating multiple configurations, which CoT lacks. giffmana counters that CoTs can actually do that. A notable debate among prominent researchers on the nature of inference-time compute and search.
Related event: LeCun vs Fleuret: Is Chain-of-Thought Really Search?(13 posts)→
More from AGI Musings
- Terence Tao on AI proofs: clever ideas, but no humans left to explain them — burny_tech · 2026-10-10
- PhD advisor's context-switching skill maps directly to managing AI agents — Michael_J_Black · 2026-10-10
- Philosophers clash over a taxonomy of consciousness theories, from dualism to IIT — burny_tech · 2026-10-10
- Consciousness theories roasted in one thread: functionalism, IIT dismissals and solipsism — burny_tech · 2026-10-10
- Ex-Cohere researcher mocks 'strategic accounting innovation' at Bay Area AI labs — suchenzang · 2026-10-10
- People Fear AI Taking Their Jobs, Not Losing the Commute and Meetings — rand_longevity · 2026-10-10