rotative
the decoder · October 6, 2026

Researchers turn LeCun's JEPA into universal world model from physics to biology

researchers extend yann lecun's jepa ai into a universal world model that works across fields like physics, robotics, and medicine, achieving big gains in dynamics tests. this could enable more accurate predictions across multiple domains.

read in 简体中文 · 繁體中文

researchers led by phai labs, with collaborators from stanford, oxford, and princeton, have expanded yann lecun's jepa architecture into jepa-anything, a universal world model that works across seven different fields, including physics, robotics, and medicine. the model breaks future states into multiple partial predictions, allowing it to pick up patterns that standard models miss. jepa-anything achieved significant gains in dynamics tests, with a 35 percent drop in prediction error in a simplified pong environment and a nearly 50 percent drop in error on the burgers equation, a common fluid dynamics benchmark. the model also identified a liver cancer treatment candidate that killed more tumor cells in lab samples and mice than either component alone. this development could enable more accurate predictions and applications across multiple domains, including medicine and robotics.

read at source ↗

← rotative