JEPA-Anything proposes one world-modeling framework across vision, biology, weather and more
gekobraa · x · 2026-09-21
A new paper introduces JEPA-Anything, a domain-agnostic world-modeling framework built on Orthogonal Predictive Factorization (OPF): a single JEPA latent state is decomposed into complementary factors that predict different parts of a system, then recombined in a shared predictive design.
- Evaluated across seven domains: vision, biology, clinical trajectories, control, molecular dynamics, physical fields, and weather.
- Improves reported metrics on all 10 matched dynamics tasks vs matched JEPA baselines; cuts single-intervention prediction error on Interventional Pong by 34.8%.
- Lowest one-step and 100-step molecular rollout errors among compared methods in all four systems.
- Latent orbital modes recover Kepler's scaling exponent (fitted slope -1.4991).
- A factor-nominated biological intervention gained experimental support in cell co-cultures, patient-derived organoids, tumor fragments and mice.
More from Research
- AI slop rejected papers will be endlessly resubmitted, warns researcher on record submission volume — menhguin · 2026-09-21
- Should LLMs be first-pass reviewers for every scientific paper? Researchers say yes — anshulkundaje · 2026-09-21
- Mystery model Jev spawns 6 open-source clones in 2 days, each with a different architecture — IgorCarron · 2026-09-21
- Framingham and TCGA were pivotal without AI; models in the loop could boost data generation — anshulkundaje · 2026-09-21
- Schulman's goal-driven research advice sparks debate on method-driven research traps — mattturck · 2026-09-21
- China releases first-round post-quantum crypto candidates across three algorithm categories — matthew_d_green · 2026-09-21