1 paper · 1 filter
Pengfei Zhou, Hexin Wang, Zhengfeiyang Zhang +5
A common strategy for scaling world models is to train on more crawled video with more compute. We argue that this strategy is inefficient: scaling world models also requires a rec…