Glossary
World model
A model that learns to predict how an environment evolves, often trained on long, continuous real-world video.
The training signal is prediction rather than labels: given what happened so far, forecast the next frames or states. That rewards long, continuous, unedited video with consistent physics — driving runs, workshop sessions, first-person daily life. Heavily edited footage is less useful because cuts break the temporal continuity the model is trying to learn.
Why it matters
The long continuous takes an editor would trim are precisely what world-model teams look for.