Cube-solving policy hits 21.2 moves with 525k-width beam search, 20% above optimal
haydendevs · x · 2026-11-13
An experimenter reports cube-solving results: a perfect solver averages 18 moves; their policy-only network is just 10 moves away without any test-time search. Running the same weights with a 100k-width beam search brought it to 21.8, and 525k width to 21.2 moves per solve — about 20% longer than optimal, likely the ceiling for this network.
Related event: Policy Network Solves Rubik's Cube Within 10 Moves of Optimal(2 posts)→
More from Research
- Neurons as an Interpretable Basis: technical follow-up to Already Clean — TemetN · 2026-11-13
- NextLat releases final paper revision: multi-token belief states, Mamba gains — JohnCLangford · 2026-11-13
- Genomics x AI community blog launches with AlphaGenome tutorials and more — anshulkundaje · 2026-11-13
- CVPR'27 paper deadline 4 days away: AI conference countdown — AlexTensor · 2026-11-13
- Each 10x in compute buys roughly 10x longer AI time horizons, elasticity near 1 — ben_j_todd · 2026-11-13
- NeuroAI work clarifies CNN vs transformer brain models — transformers win in interesting ways — dyamins · 2026-11-13