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

arXiv: https://arxiv.org/abs/2503.15477 · HTML: https://mlsanity.com/paper/2503.15477/

## Key facts

- Cited by 73 papers in the map (data through 14 Oct 2026).
- Placed in Reward modeling and RLHF, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 35 of its 95 references that are in the map.
- First posted on arXiv 19 Mar 2025.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Reward modeling and RLHF](https://mlsanity.com/alignment/reward-modeling-and-rlhf/)

## Builds on (most cited first, 30 of 35)

- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2204.05862/) (2022)
- [HuggingFace's Transformers: State-of-the-art Natural Language Processing](https://mlsanity.com/paper/1910.03771/) (2019)
- [Fine-Tuning Language Models from Human Preferences](https://mlsanity.com/paper/1909.08593/) (2019)
- [WebGPT: Browser-assisted question-answering with human feedback](https://mlsanity.com/paper/2112.09332/) (2021)
- [Concrete Problems in AI Safety](https://mlsanity.com/paper/1606.06565/) (2016)
- [Tulu 3: Pushing Frontiers in Open Language Model Post-Training](https://mlsanity.com/paper/2411.15124/) (2024)
- [RAFT: Reward rAnked FineTuning for Generative Foundation Model Alignment](https://mlsanity.com/paper/2304.06767/) (2023)
- [SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training](https://mlsanity.com/paper/2501.17161/) (2025)
- [Zephyr: Direct Distillation of LM Alignment](https://mlsanity.com/paper/2310.16944/) (2023)
- [Reinforced Self-Training (ReST) for Language Modeling](https://mlsanity.com/paper/2308.08998/) (2023)
- [Interpretable Preferences via Multi-Objective Reward Modeling and Mixture-of-Experts](https://mlsanity.com/paper/2406.12845/) (2024)
- [Is DPO Superior to PPO for LLM Alignment? A Comprehensive Study](https://mlsanity.com/paper/2404.10719/) (2024)
- [Secrets of RLHF in Large Language Models Part I: PPO](https://mlsanity.com/paper/2307.04964/) (2023)
- [Neural Policy Gradient Methods: Global Optimality and Rates of Convergence](https://mlsanity.com/paper/1909.01150/) (2019)
- [Understanding the performance gap between online and offline alignment algorithms](https://mlsanity.com/paper/2405.08448/) (2024)
- [Back to Basics: Revisiting REINFORCE Style Optimization for Learning from Human Feedback in LLMs](https://mlsanity.com/paper/2402.14740/) (2024)
- [Active Preference Optimization for Sample Efficient RLHF](https://mlsanity.com/paper/2402.10500/) (2024)
- [Provable Offline Preference-Based Reinforcement Learning](https://mlsanity.com/paper/2305.14816/) (2023)
- [Improving Reinforcement Learning from Human Feedback with Efficient Reward Model Ensemble](https://mlsanity.com/paper/2401.16635/) (2024)
- [Reinforcement Learning from Human Feedback with Active Queries](https://mlsanity.com/paper/2402.09401/) (2024)
- [Ray Interference: a Source of Plateaus in Deep Reinforcement Learning](https://mlsanity.com/paper/1904.11455/) (2019)
- [InfAlign: Inference-aware language model alignment](https://mlsanity.com/paper/2412.19792/) (2024)
- [Optimal Design for Reward Modeling in RLHF](https://mlsanity.com/paper/2410.17055/) (2024)
- [The Accuracy Paradox in RLHF: When Better Reward Models Don't Yield Better Language Models](https://mlsanity.com/paper/2410.06554/) (2024)
- [Stronger Models are NOT Stronger Teachers for Instruction Tuning](https://mlsanity.com/paper/2411.07133/) (2024)
- [Neural Replicator Dynamics](https://mlsanity.com/paper/1906.00190/) (2019)

## Built on by (most cited first, 30 of 73)

- [Reinforcement Learning for Reasoning in Large Language Models with One Training Example](https://mlsanity.com/paper/2504.20571/) (2025)
- [TTRL: Test-Time Reinforcement Learning](https://mlsanity.com/paper/2504.16084/) (2025)
- [RewardBench 2: Advancing Reward Model Evaluation](https://mlsanity.com/paper/2506.01937/) (2025)
- [Checklists Are Better Than Reward Models For Aligning Language Models](https://mlsanity.com/paper/2507.18624/) (2025)
- [Retaining by Doing: The Role of On-Policy Data in Mitigating Forgetting](https://mlsanity.com/paper/2510.18874/) (2025)
- [Not All Rollouts are Useful: Down-Sampling Rollouts in LLM Reinforcement Learning](https://mlsanity.com/paper/2504.13818/) (2025)
- [RewardDance: Reward Scaling in Visual Generation](https://mlsanity.com/paper/2509.08826/) (2025)
- [EditScore: Unlocking Online RL for Image Editing via High-Fidelity Reward Modeling](https://mlsanity.com/paper/2509.23909/) (2025)
- [RLVE: Scaling Up Reinforcement Learning for Language Models with Adaptive Verifiable Environments](https://mlsanity.com/paper/2511.07317/) (2025)
- [Consistent Paths Lead to Truth: Self-Rewarding Reinforcement Learning for LLM Reasoning](https://mlsanity.com/paper/2506.08745/) (2025)
- [RePO: Replay-Enhanced Policy Optimization](https://mlsanity.com/paper/2506.09340/) (2025)
- [MMR1: Enhancing Multimodal Reasoning with Variance-Aware Sampling and Open Resources](https://mlsanity.com/paper/2509.21268/) (2025)
- [ETTRL: Balancing Exploration and Exploitation in LLM Test-Time Reinforcement Learning Via Entropy Mechanism](https://mlsanity.com/paper/2508.11356/) (2025)
- [Cultivating Pluralism In Algorithmic Monoculture: The Community Alignment Dataset](https://mlsanity.com/paper/2507.09650/) (2025)
- [100 Days After DeepSeek-R1: A Survey on Replication Studies and More Directions for Reasoning Language Models](https://mlsanity.com/paper/2505.00551/) (2025)
- [RewardAnything: Generalizable Principle-Following Reward Models](https://mlsanity.com/paper/2506.03637/) (2025)
- [VL-Cogito: Progressive Curriculum Reinforcement Learning for Advanced Multimodal Reasoning](https://mlsanity.com/paper/2507.22607/) (2025)
- [Discriminative Policy Optimization for Token-Level Reward Models](https://mlsanity.com/paper/2505.23363/) (2025)
- [The Best of N Worlds: Aligning Reinforcement Learning with Best-of-N Sampling via max@k Optimisation](https://mlsanity.com/paper/2510.23393/) (2025)
- [Pre-Trained Policy Discriminators are General Reward Models](https://mlsanity.com/paper/2507.05197/) (2025)
- [Sailing by the Stars: A Survey on Reward Models and Learning Strategies for Learning from Rewards](https://mlsanity.com/paper/2505.02686/) (2025)
- [Small Generalizable Prompt Predictive Models Can Steer Efficient RL Post-Training of Large Reasoning Models](https://mlsanity.com/paper/2602.01970/) (2026)
- [Reinforcement Learning with Inverse Rewards for World Model Post-training](https://mlsanity.com/paper/2509.23958/) (2025)
- [Improving Generalization in Intent Detection: GRPO with Reward-Based Curriculum Sampling](https://mlsanity.com/paper/2504.13592/) (2025)
- [Reshaping Reasoning in LLMs: A Theoretical Analysis of RL Training Dynamics through Pattern Selection](https://mlsanity.com/paper/2506.04695/) (2025)
- [Bradley-Terry and Multi-Objective Reward Modeling Are Complementary](https://mlsanity.com/paper/2507.07375/) (2025)
- [Understanding the Performance Gap in Preference Learning: A Dichotomy of RLHF and DPO](https://mlsanity.com/paper/2505.19770/) (2025)
- [SuperFlow: Training Flow Matching Models with RL on the Fly](https://mlsanity.com/paper/2512.17951/) (2025)
- [LearnAlign: Data Selection for LLM Reinforcement Learning with Improved Gradient Alignment](https://mlsanity.com/paper/2506.11480/) (2025)
- [Behavior Injection: Preparing Language Models for Reinforcement Learning](https://mlsanity.com/paper/2505.18917/) (2025)

## Abstract (arXiv)

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.