nyu

Sub-JEPA improves LeWorldModel by applying Gaussian regularization in frozen random orthogonal subspaces, consistently outperforming the original on benchmarks with up to +10.7 percentage points improvement.

Yann LeCun closed a $1.03B funding round for AMI Labs, followed by a new paper from his NYU collaborators introducing LeWorldModel, a lightweight JEPA-based world model that runs on a single GPU.