Verification board

New results arrive faster than anyone can referee them: preprints, lab announcements, AI-generated proofs. Each claim here is broken into concrete checks (rerun this computation, match this Lean statement to the paper, find an independent dataset). Pick one, check it, and explain what you found. Certified experts settle each check against the evidence.

Put a claim on the board

claims on the board

waiting for verification

open checks to pick up

verifications submitted

All Unverified · 13 Under review · 13 Partly verified · 4 Verified · 3 Refuted · 3 Retracted · 2

All areas Algorithms & Complexity Computational Chemistry Computational Medicine GPU Kernels & Systems Machine Learning Mathematics Other Physics

UnverifiedPostedComputationReinforcement Learning

RL compute for LLMs follows predictable sigmoidal curves that extrapolate to a 100,000 GPU-hour run

Khatri et al., arXiv preprint · Oct 15, 2025

On Do RL compute-performance curves for LLMs extrapolate across models, domains and recipes?

✓ 0 passed✗ 0 failed○ 1 open0 verifications

UnverifiedPostedImplementationReinforcement Learning

An agent trained only inside a learned world model from offline data obtains Minecraft diamonds

Hafner et al. (Dreamer 4), arXiv preprint · Sep 29, 2025

Independent reproduction and improvement of offline-only Minecraft diamond collection with a world model