<ul data-eligibleForWebStory="true"><li>Meta introduced the V-JEPA 2 AI model aimed at helping AI agents comprehend their environment.</li><li>V-JEPA 2 is an evolution of last year's V-JEPA model trained on 1 million hours of video to aid robots in understanding physical world concepts like gravity.</li><li>The model grasps common sense connections seen in small children and animals, enabling predictions of consequences like rebounding a ball or moving cooked eggs with a spatula.</li><li>Meta's V-JEPA 2 outperforms Nvidia's Cosmos model in speed (30 times faster), though evaluation criteria may differ.</li><li>Meta envisions world models revolutionizing robotics by empowering AI agents to execute physical tasks with minimal robotic training data.</li></ul>

Meta’s V-JEPA 2 model teaches AI to understand its surroundings

Discover more