What Makes a Reward Model a Good Teacher? An Optimization Perspective

NeurIPSSpotlight2025

Authors: Noam Razin, Zixuan Wang, Hubert Strauss, Stanley Wei, Jason D. Lee, Sanjeev Arora
Venue: NeurIPS 2025
Track: Spotlight

TL;DR

The success of Reinforcement Learning from Human Feedback (RLHF) critically depends on the quality of the reward model…

Opening excerpt from the authors’ abstract. source

Read the paper

OpenReview

Topics

reinforcement learning optimization rlhf

← All NeurIPS 2025 Spotlight papers · Browse the whole archive