THE COLLECTION
EVERY SIGNAL, SORTED.
Search the whole shelf by topic, tool, term, or plain old curiosity.
SHOWING 1 OF 517 STORIES
ISSUE 505EVENING
France, Kyutai, World Models, Video Autoencoders, Robotics, Autonomous Driving, Human Manipulation, Interactive Simulation, Latency, Open Research and Model Evaluation
Kyutai made world models react faster. Measure the control lag
Kyutai released causal video autoencoders and four action-conditioned world models that can generate interactive scenes at 20 frames per second on one H100. The useful number is not frame rate alone. It is the full delay from a person or machine issuing an action to the displayed world visibly doing the right thing.