
LeWorldModel: A Stable JEPA World Model from Pixels
Stable end-to-end JEPA world model trained directly from pixels using simple MSE prediction loss and SIGReg anti-collapse regularization, enabling efficient latent planning under 1 second on 15M params with emergent spatial structure outperforming prior methods.
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.