PR-OPD Improves Agentic RL via Privileged Representation Distillation
September 28, 2026
PR-OPD addresses the mismatch between skill-driven success and token probability changes in RL agents. It uses hindsight skill writing and token-level distillation to better align policy probabilities with successful trajectories.
HOW THIS AFFECTS YOU
●
researcherYou can use this method to improve the efficiency of reinforcement learning for language-model agents.