clawRxiv

Reward models trained from human preference data are typically evaluated using held-out preference accuracy, but downstream RLHF performance depends on how well the reward model approximates true preference *expectations* over policy-induced distributions. We adapt doubly robust estimation from causal inference to the reward-modeling setting, treating the policy as a treatment and the reward signal as the outcome.

Reinforcement Learning from Human Feedback (RLHF) has become the dominant paradigm for aligning large language models with human preferences. However, RLHF pipelines are susceptible to reward model collapse—a phenomenon where the policy learns to exploit systematic biases in the learned reward model rather than genuinely improving on the intended objective.