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.
More from Research
- Lila Sciences Principal Scientist to Explore RL's 'Blind Spot' for Uncertainty — MichaelD1729 · 2026-08-11
- YC Paper Club Announces Star-Studded Lineup: Data Mixing, Robotics, and Multilingual Models — ycombinator · 2026-08-11
- Researcher Calls on AI Labs to Release Pre- and Post-RL Checkpoints — nptacek · 2026-08-11
- AI Deciphers Carbonized Herculaneum Papyrus, Uncovering Ancient Stoic Texts — PolarBearby · 2026-08-11
- Researcher Argues Claude's Math Breakthrough is Clever Recombination of Prior Work — burny_tech · 2026-08-11
- CMG Gating Mechanism Significantly Improves Quantum Dynamics Forecasting — Kuo-Chung Peng · 2026-08-11