Meta rolled out a new AI "world model" called V-JEPA 2 that understands 3D environments and physical object movements. The AI can predict real-world physics, like a ball rolling off a table will fall, or that hidden objects still exist.
The open-source V-JEPA 2 builds an internal simulation of reality, enabling machines to learn, plan, and make human-like decisions. Meta highlighted potential uses in delivery robots and self-driving cars that need to navigate complex surroundings in real time.
Instead of relying on huge labeled datasets or video footage, V-JEPA 2 reasons in a simplified "latent" space to figure out how objects move and interact. 
Meta’s chief AI scientist Yann LeCunn said during the Viva Tech conference in Paris:
"Allowing machines to understand the physical world is very different from allowing them to understand language."
"A world model is like an abstract digital twin of reality that an AI can reference to understand the world and predict consequences of its actions and therefore it would be able to plan a course of action to accomplish a given task."
Meta faces stiff AI competition from OpenAI, Microsoft, and Google. The company is investing $14 billion into AI startup Scale AI and plans to hire its CEO Alexandr Wang to push its AI strategy further.
The launch follows moves by other industry players. Google’s DeepMind is developing a similar world model called Genie, which simulates 3D environments in real time. Meanwhile, AI researcher Fei-Fei Li raised $230 million for World Labs to build "large world models" that grasp the physical world’s structure.
Meta aims to take on the next big AI challenge — understanding reality, not just language.
