GPT-5.6 Pro improves lower bound for Moser's worm problem to 0.2374
zero0_one1 · reddit · 2026-09-04
A Reddit user reports a new lower bound for Moser's convex worm problem using the ProofAtlas.ai harness with GPT-5.6 Pro: every convex universal cover for unit-length planar curves has area greater than 0.2374, improving the 2013 best of 0.232239. The 60-year-old problem asks for the smallest-area convex region containing any planar curve of length one; the smallest known cover (Wichiramala & Panraksa, 2026) is about 0.260956, leaving a gap under 0.024. The proof is Lean-formalized and published on ProofAtlas.ai.
More from Models
- Opus 5.5 tops OpenRouter spending in 20 of 29 task categories — mmurph · 2026-10-21
- Meta team builds open-source 30B LLM from scratch in months, hails 'second era' — TimDarcet · 2026-10-21
- DFlash drafter for Muse Glimmer-30B: 5 layers, blockwise 16-token spec decoding — TimDarcet · 2026-10-21
- EvoPress quantization shrinks 30B model from 56GB to 19GB at near-bf16 accuracy — TimDarcet · 2026-10-21
- Radar chart: Muse Glimmer-30B benchmarked against Gemma4-31B and Qwen3.6-27B — TimDarcet · 2026-10-21
- Privacy evals: Muse Glimmer-30B leaks less than Qwen3.6-27B on CIMemories — TimDarcet · 2026-10-21