Selecting Long-Horizon Trajectories for Reliable and Efficient Terminal-Agent Training
Terminal agents are commonly trained by imitating long teacher trajectories, yet how much of each trajectory to supervise remains unexplored. We study the \emph{supervision horizon}, the number of trajectory tokens retained for training, and show that it is a key design axis for reliability and cost. Reliability improves with longer horizons but saturates: on Terminal-Bench, a 12K-token horizon…
We haven't written up this one. arXiv cs.AI has the full story — the link below goes straight to it.