Meta Unveils Video-Trained World Model V-JEPA 2

Meta has unveiled V-JEPA 2, a world model trained on video data, introduced by Chief AI Scientist Yann LeCun. With 1.2 billion parameters, the model is built on the Joint Embedding Predictive Architecture (JEPA) and features advanced environmental understanding and prediction capabilities, enabling zero-shot planning and robot control in novel environments. Trained through self-supervised learning without the need for human annotations, V-JEPA 2 can be applied in areas such as assistive technology, mixed reality guidance, and personalized education.