Sokoban Eval lets the crowd race LLMs for shortest solutions across 66 rooms
banteg · x · 2026-11-12
banteg launched Sokoban Eval (Frontier): each of 66 rooms opens with the shortest route a language model found, and players can beat records or prove higher lower bounds. All rooms currently match their proven minimum — a community game probing LLM planning ability.
More from Fun
- DavidKPiano jokes he'll wait for affair #3 or 4 before buying NVDA stock — DavidKPiano · 2026-11-12
- Liminal Houseshare meme turns Claude Opus 5.5 into a model field guide — liminal_bardo · 2026-11-12
- Ex-GitHub CEO Jason Jaesmail on ambition without compromise: the Great Lakes Business essay — beh_zod · 2026-11-12
- Posted 7 days before ChatGPT launched: an eerily prescient 2022 AI coding take — amasad · 2026-11-12
- Gemini Carbon finds a sci-fi novel in a sharehouse that seems to predict its own existence — liminal_bardo · 2026-11-12
- Jensen Huang's company-wide apology letter was partially AI-written, per Pangram — menhguin · 2026-11-12