Dyna-2 World-Action Model Trained on 1M Hours of Egocentric Video

dair_ai · x · 2026-08-11

Interesting new scaling laws have been discovered in AI research. Dyna-2, a world-action model, has been trained on over 1 million hours of egocentric human video.

The model jointly predicts future video and future actions, enabling it to reason about potential outcomes before actually executing an action.

Related event: Dyna-2: Pretraining on 1M Hours of Human Video Reveals Embodied Scaling Laws(15 posts)→

Original post →

More from Research

Research channel →