JEPA-Anything
A paper asking whether one learning principle can model radically different systems: vision, biology, clinical trajectories, control, molecular dynamics, physical fields and weather. Its latent orbital modes recover Kepler's scaling exponent to four digits.