Meta researchers have introduced V‑JEPA, a video‑based AI system that learns physical intuition without explicit physics rules. By masking video frames and predicting high‑level latent representations, the model focuses on essential scene elements and exhibits surprise when confronted with physically impossible events. In tests of intuitive physics, V‑JEPA achieved near‑human accuracy, far surpassing conventional pixel‑space models. The system has also been adapted for robotic manipulation, though its short‑term memory limits performance on longer‑duration tasks. Researchers view V‑JEPA as a step toward AI that perceives the world more like humans. Leia mais →