Meta today introduced V-JEPA 2, a 1.2-billion-parameter world model trained primarily on video to support understanding, prediction, and planning in robotic systems. Built on the Joint Embedding ...