sc94597 2 hours ago
Wonder if LeWM has anything to do with it. Sora was OpenAI's attempt at world modeling, but autoregressive generative models haven't shown to be as efficient/good as other architectures.
| With ~15M parameters trainable on a single GPU in a few hours, LeWM plans up to 48× faster than foundation-model-based world models while remaining competitive across diverse 2D and 3D control tasks. Beyond control, we show that LeWM's latent space encodes meaningful physical structure through probing of physical quantities. Surprise evaluation confirms that the model reliably detects physically implausible events. |
