Gains and Collapse in On-Policy Distillation:A Reinforcement Learning Perspective Paper • 2610.03185 • Published Oct 2 • 30