What Makes a Reward Model a Good Teacher? An Optimization Perspective
This addresses a fundamental limitation in RLHF for AI alignment, revealing that reward models must balance accuracy and variance for effective teaching, which is incremental but crucial for improving training efficiency.
The paper tackles the problem that reward model accuracy alone does not guarantee efficient optimization in RLHF, showing that low reward variance leads to a flat objective landscape and slower training, with experiments on models up to 8B parameters confirming this trade-off.
The success of Reinforcement Learning from Human Feedback (RLHF) critically depends on the quality of the reward model. However, while this quality is primarily evaluated through accuracy, it remains unclear whether accuracy fully captures what makes a reward model an effective teacher. We address this question from an optimization perspective. First, we prove that regardless of how accurate a reward model is, if it induces low reward variance, then the RLHF objective suffers from a flat landscape. Consequently, even a perfectly accurate reward model can lead to extremely slow optimization, underperforming less accurate models that induce higher reward variance. We additionally show that a reward model that works well for one language model can induce low reward variance, and thus a flat objective landscape, for another. These results establish a fundamental limitation of evaluating reward models solely based on accuracy or independently of the language model they guide. Experiments using models of up to 8B parameters corroborate our theory, demonstrating the interplay between reward variance, accuracy, and reward maximization rate. Overall, our findings highlight that beyond accuracy, a reward model needs to induce sufficient variance for efficient~optimization.