Meta on Wednesday open new V-jepa 2 AI Model, “The World Model” designed to help you do what the world knows.
V-jepa 2 is an extension of V-jepa A model that is released in the last year, which is trained for more than one million hours. This training data is supposed to help other robots or other Agent ADI used in the physical world, understanding and predict the concept like gravity will be in the next.
This is a type of a common connection with the minor and animal developed – if you play with your dog, or how it can produce the race on the ground, and it is not in the ball.
Meta describes an example where the robot can be faced, such as, a point of viewing plate and spatula and walks into the stove with the cooking egg. AI can predict that the next action will use a spatula to move eggs to the plate.
According to Meta, V-Jepa 2 Age faster than NVIDIA Cosmos Models, which also try to increase the interest associated with the physical world. However, meta may evaluate your own model according to different benchmarks than nvidia.
“We believe the world model will pursue the new era to make a real world Agent to help your job and physical task without requires robotical training data,” explained Head of AI mom Ain scientist in the video.
