1 paper
Pardis Taghavi, Tingyu Guo, Jonas Lossner +2
Chunk-autoregressive video world models typically generate each chunk under one action. When an action changes during sampling, waiting until the next chunk delays the response, wh…