MC-PhysBench: leading video world models break physics within a few frames
CatAstro_Piyush · x · 2026-10-09
Midcentury AI tested leading video world models on their ability to predict simple physics and found most fail within a few frames.
They're releasing MC-PhysBench, a new benchmark measuring how long a model's output stays physically valid: 15+ models, 9 scenes, 6,000+ videos scored, with 7 failure checks. It's the first eval to make physical-validity duration the core metric, probing how far video world models are from real-world simulation.
More from Research
- AI-designed nanoparticles stabilize mRNA vaccines at high temperatures, Nature Biotech reports — emmanuelvivier · 2026-10-09
- LLM agent polarization study from Tsinghua-UChicago team published in Nature Communications — profjamesevans · 2026-10-09
- Forced contact moves LLM agent opinions just 2%, polarization study finds — profjamesevans · 2026-10-09
- LLM agents build human-like social networks — and polarize anyway — profjamesevans · 2026-10-09
- Mathematician downplays Navier-Stokes 'mistranslation' uproar: no doubt about the proof — basedjensen · 2026-10-09
- Prima Mente Launches Alzheimer's AI Challenge With 150M-Cell Perturbation Atlas — katieruthmishra · 2026-10-09