Meta’s V-Jepa 2 Model, I have to understand their surroundings

Meta the Wednesday has submitted their new V-2PA 2 The Model Ai, a “World Model” that is designed to help the agents I understand the world around them.

V-jepa 2 is an extension of the V-jepa Meta Model released last year, which has been trained on more than one million hours of video hours. This training data is supposed to robot or other agents aven in the physical world, understanding and predict as gravity impacts me in a sequence.

Here are common sense connection ships that children get the little ones for the brigis – how it has to be rendered from here, and not where the dance is that the accurate moment.

Meta rapping you where a robot can be confronted, eg the point of view of a plate and a spatula and to stay at the cooked egg. I can predict that a person next the next would you use the spatula to move eggs to the plate.

According to Meta, V-JPA 2 is 30X faster than nvidia CSMOS pattern, that also tries to improve intelligence related to the physical world. However, meta can be assessing their own patterns according to different benchmarks than nvidia.

“We believe the world’s patterns will be a new way to robotics, allow the avenue of the real world to help the physical jobs without needing astronomic training data” Explained The member’s leader ai skiientist yann yann in a video.