ASH: Agents that Self-Hone in Long-Horizon Worlds
ASH is introduced, an agentic system that learns a long-horizon policy from unlabeled, noisy internet video, without reward shaping or expert annotation, and it is demonstrated that self-improving agents are a scalable recipe for learning in long-horizon worlds.
Benjamin Schneider, Xavier Schneider, Victor Zhong et al.
· 0 citations