Meta just dropped V-JEPA 2, an AI model that helps machines actually "get" how the physical world works. By learning from tons of videos—think stuff like gravity or objects moving—it could make robots and self-driving cars way smarter.
V-JEPA 2 is 30 times faster than NVIDIA's Cosmos model
V-JEPA 2 isn't just smart; it's quick. It figures out what's happening in videos and predicts outcomes up to 30 times faster than NVIDIA's Cosmos model. That speed could be a game-changer for tech that needs to react in real time.
A leap forward for AI learning
Unlike older models that rely on pixel-by-pixel predictions, V-JEPA 2 learns on its own from huge amounts of data—no human hand-holding needed. It represents a big leap for AI learning.