Claude Solves Open Physics Problem, Crooks Predicts Major Academic Shift
新智元 · wechat · 2026-08-28
Physicist Gavin E. Crooks (a founder of non-equilibrium statistical mechanics) gave Claude an open problem regarding the entropy production constraints of the Detailed Fluctuation Theorem (DFT). Within days, Claude completely solved the problem, providing a unified geometric image (convex body) for all known local results and writing the paper from abstract to conclusion.
Claude's Key Insight:
- Claude identified that every distribution satisfying DFT corresponds to a 'gap distribution', and any DFT distribution is a unique mixture of two basic outcome distributions.
- It proved that the joint reachable region of statistics is a convex body with precise characterization at every order: a sharp lower bound but no upper bound.
- This finding unifies all previously published DFT bounds as low-dimensional projections of this convex body, explaining how a single distribution can saturate multiple bounds simultaneously.
Industry Impact:
- Crooks noted that this is not just a thermodynamics result but marks AI's shift from 'solving known human problems' to 'exploring unknown human problems.'
- He predicts physics will enter an AI-accelerated era like mathematics, with academia facing massive impacts on both knowledge advancement and education/training roles.
More from AGI Musings
- Why Did GPT-5.5 Love Goblins? Unpredictable AI Alignment Sparks Debate — louisvarge · 2026-08-28
- Opinion: Pause Pretraining to Focus on Controlling Inner Model Drives — louisvarge · 2026-08-28
- Thought experiment: Pause pretraining to focus on controlling inner drives — louisvarge · 2026-08-28
- Judea Pearl: gene-IQ debate can be settled the same way as smoking-cancer was — yudapearl · 2026-08-28
- AI Reliability Lags 4-10x Behind Capabilities; 90% Accuracy Is Catastrophic — sayashk · 2026-08-28
- Google DeepMind uses Teamwork multi-agent framework for math breakthroughs — algo_diver · 2026-08-28