# Learning a Pessimistic Reward Model in RLHF

arXiv: https://arxiv.org/abs/2505.20556 · HTML: https://mlsanity.com/paper/2505.20556/

## Key facts

- Cited by 5 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 43 of its 51 references that are in the map.
- First posted on arXiv 26 May 2025.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Reward modeling and RLHF](https://mlsanity.com/alignment/reward-modeling-and-rlhf/)
- Offline RL and imitation learning from data / [Offline RL algorithms](https://mlsanity.com/offlinerl/offline-rl-algorithms/)

## Builds on (most cited first, 30 of 43)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [LLaMA: Open and Efficient Foundation Language Models](https://mlsanity.com/paper/2302.13971/) (2023)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Deep reinforcement learning from human preferences](https://mlsanity.com/paper/1706.03741/) (2017)
- [Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2204.05862/) (2022)
- [Constitutional AI: Harmlessness from AI Feedback](https://mlsanity.com/paper/2212.08073/) (2022)
- [Learning to summarize from human feedback](https://mlsanity.com/paper/2009.01325/) (2020)
- [Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems](https://mlsanity.com/paper/2005.01643/) (2020)
- [Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling](https://mlsanity.com/paper/2304.01373/) (2023)
- [Scaling Laws for Reward Model Overoptimization](https://mlsanity.com/paper/2210.10760/) (2022)
- [Iterative Preference Learning from Human Feedback: Bridging Theory and Practice for RLHF under KL-Constraint](https://mlsanity.com/paper/2312.11456/) (2023)
- [Statistical Rejection Sampling Improves Preference Optimization](https://mlsanity.com/paper/2309.06657/) (2023)
- [RLHF Workflow: From Reward Modeling to Online RLHF](https://mlsanity.com/paper/2405.07863/) (2024)
- [Helping or Herding? Reward Model Ensembles Mitigate but do not Eliminate Reward Hacking](https://mlsanity.com/paper/2312.09244/) (2023)
- [Direct Nash Optimization: Teaching Language Models to Self-Improve with General Preferences](https://mlsanity.com/paper/2404.03715/) (2024)
- [Generalized Preference Optimization: A Unified Approach to Offline Alignment](https://mlsanity.com/paper/2402.05749/) (2024)
- [WARM: On the Benefits of Weight Averaged Reward Models](https://mlsanity.com/paper/2401.12187/) (2024)
- [A Minimalist Approach to LLM Reasoning: from Rejection Sampling to Reinforce](https://mlsanity.com/paper/2504.11343/) (2025)
- [Scaling Laws for Reward Model Overoptimization in Direct Alignment Algorithms](https://mlsanity.com/paper/2406.02900/) (2024)
- [Theoretical guarantees on the best-of-n alignment policy](https://mlsanity.com/paper/2401.01879/) (2024)
- [Adversarially Trained Actor Critic for Offline Reinforcement Learning](https://mlsanity.com/paper/2202.02446/) (2022)
- [Provably Mitigating Overoptimization in RLHF: Your SFT Loss is Implicitly an Adversarial Regularizer](https://mlsanity.com/paper/2405.16436/) (2024)
- [Exploratory Preference Optimization: Harnessing Implicit Q*-Approximation for Sample-Efficient RLHF](https://mlsanity.com/paper/2405.21046/) (2024)
- [Is Best-of-N the Best of Them? Coverage, Scaling, and Optimality in Inference-Time Alignment](https://mlsanity.com/paper/2503.21878/) (2025)
- [Correcting the Mythos of KL-Regularization: Direct Alignment without Overoptimization via Chi-Squared Preference Optimization](https://mlsanity.com/paper/2407.13399/) (2024)
- [RRM: Robust Reward Model Training Mitigates Reward Hacking](https://mlsanity.com/paper/2409.13156/) (2024)
- [Value-Incentivized Preference Optimization: A Unified Approach to Online and Offline RLHF](https://mlsanity.com/paper/2405.19320/) (2024)
- [Robust Preference Optimization through Reward Model Distillation](https://mlsanity.com/paper/2405.19316/) (2024)
- [Self-Exploring Language Models: Active Preference Elicitation for Online Alignment](https://mlsanity.com/paper/2405.19332/) (2024)

## Built on by (most cited first, 5 of 5)

- [Learning from Failures: Understanding LLM Alignment through Failure-Aware Inverse RL](https://mlsanity.com/paper/2510.06092/) (2025)
- [Golden Handcuffs make safer AI agents](https://mlsanity.com/paper/2604.13609/) (2026)
- [Themis: Training Robust Multilingual Code Reward Models for Flexible Multi-Criteria Scoring](https://mlsanity.com/paper/2605.00754/) (2026)
- [Circuit-Aware Reward Training: A Mechanistic Framework for Longtail Robustness in RLHF](https://mlsanity.com/paper/2509.24713/) (2025)
- [Wasserstein Distributionally Robust Regret Optimization for Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2605.00155/) (2026)

## Abstract (arXiv)

This work proposes `PET', a novel pessimistic reward fine-tuning method, to learn a pessimistic reward model robust against reward hacking in offline reinforcement learning from human feedback (RLHF). Traditional reward modeling techniques in RLHF train an imperfect reward model, on which a KL regularization plays a pivotal role in mitigating reward hacking when optimizing a policy. Such an intuition-based method still suffers from reward hacking, and the policies with large KL divergence from the dataset distribution are excluded during learning. In contrast, we show that when optimizing a policy on a pessimistic reward model fine-tuned through PET, reward hacking can be prevented without relying on any regularization. We test our methods on the standard TL;DR summarization dataset. We find that one can learn a high-quality policy on our pessimistic reward without using any regularization. Such a policy has a high KL divergence from the dataset distribution while having high performance in practice. In summary, our work shows the feasibility of learning a pessimistic reward model against reward hacking. The agent can greedily search for the policy with a high pessimistic reward without suffering from reward hacking.