One Model, Seven Worlds

🔥 What happened A 13-author team dropped JEPA-Anything, a domain-agnostic framework that trains predictive world models across seven wildly different fields — vision, biology, clinical trajectories, control, molecular dynamics, physical fields, and weather. 💡 Why it matters One shared predictive principle replaces seven specialist models. It cuts single-intervention prediction error on Interventional Pong by 34.8% and recovers the Keplerian scaling exponent at -1.4991 from latent orbital modes. ⚡ Our take If this generalizes, domain-specific world models are yesterday's news. But "Anything" claims from arXiv abstracts have a long history of dying in the lab.