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

Wednesday unveiled Meta’s new V-JPA 2 AI model, a “world model” that is designed to help AI agents understand the world around them.

An extension of V-Zepa 2 V-jP The model that was published last year, which was trained in more than a million hours of video. This training is believed to help robots or other AI agents operate in the physical world, understanding and predicting how gravity will affect how the ideas of gravity occur after a sequence.

These are small children and animals to develop their brain as well as common sense connections – when you play with the dog, for example, the dog (optimistic) understands how to bounce a ball on the ground, or how it should run it where it should land, and the ball is not the right moment.

Meta depicts examples where a robot can face, for example, holding a plate and a spatula and walking point of view on the oven with cooked eggs. AI may predict that the next verb is very likely to use spatula to move the eggs to the plate.

According to the meta, V-JPA 2 faster than 2 nvidia Cosmos The model, which tries to enhance the intelligence related to the physical world. However, Meta can evaluate its own models according to various criteria than the Nvidia.

“We believe that the world models will introduce a new era for robotics, the amount of astronomy in real world AI agents will enable work and physical work to work without need of robotic training,” Explanation In a video, Meta chief AI scientist Ian Lucun.