# Reward Model Interpretability via Optimal and Pessimal Tokens

arXiv: https://arxiv.org/abs/2506.07326 · HTML: https://mlsanity.com/paper/2506.07326/

## Key facts

- Cited by 10 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 33 of its 57 references that are in the map.
- First posted on arXiv 8 Jun 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 33)

- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [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)
- [Universal and Transferable Adversarial Attacks on Aligned Language Models](https://mlsanity.com/paper/2307.15043/) (2023)
- [Learning to summarize from human feedback](https://mlsanity.com/paper/2009.01325/) (2020)
- [Visualizing and Understanding Convolutional Networks](https://mlsanity.com/paper/1311.2901/) (2013)
- [Gemma 2: Improving Open Language Models at a Practical Size](https://mlsanity.com/paper/2408.00118/) (2024)
- [Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference](https://mlsanity.com/paper/2403.04132/) (2024)
- [Scaling Laws for Reward Model Overoptimization](https://mlsanity.com/paper/2210.10760/) (2022)
- [Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2307.15217/) (2023)
- [AlpacaFarm: A Simulation Framework for Methods that Learn from Human Feedback](https://mlsanity.com/paper/2305.14387/) (2023)
- [Interpretable Preferences via Multi-Objective Reward Modeling and Mixture-of-Experts](https://mlsanity.com/paper/2406.12845/) (2024)
- [Skywork-Reward: Bag of Tricks for Reward Modeling in LLMs](https://mlsanity.com/paper/2410.18451/) (2024)
- [A Long Way to Go: Investigating Length Correlations in RLHF](https://mlsanity.com/paper/2310.03716/) (2023)
- [Way Off-Policy Batch Deep Reinforcement Learning of Implicit Human Preferences in Dialog](https://mlsanity.com/paper/1907.00456/) (2019)
- [A Roadmap to Pluralistic Alignment](https://mlsanity.com/paper/2402.05070/) (2024)
- [Personalizing Reinforcement Learning from Human Feedback with Variational Preference Learning](https://mlsanity.com/paper/2408.10075/) (2024)
- [Reducing Gender Bias in Abusive Language Detection](https://mlsanity.com/paper/1808.07231/) (2018)
- [The PRISM Alignment Dataset: What Participatory, Representative and Individualised Human Feedback Reveals About the Subjective and Multicultural Alignment of Large Language Models](https://mlsanity.com/paper/2404.16019/) (2024)
- [Regularizing Hidden States Enables Learning Generalizable Reward Model for LLMs](https://mlsanity.com/paper/2406.10216/) (2024)
- [Personalized Language Modeling from Personalized Human Feedback](https://mlsanity.com/paper/2402.05133/) (2024)
- [Unpacking DPO and PPO: Disentangling Best Practices for Learning from Preference Feedback](https://mlsanity.com/paper/2406.09279/) (2024)
- [Human Feedback is not Gold Standard](https://mlsanity.com/paper/2309.16349/) (2023)
- [Loose lips sink ships: Mitigating Length Bias in Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2310.05199/) (2023)
- [Uncertainty-aware Reward Model: Teaching Reward Models to Know What is Unknown](https://mlsanity.com/paper/2410.00847/) (2024)
- [Elo Uncovered: Robustness and Best Practices in Language Model Evaluation](https://mlsanity.com/paper/2311.17295/) (2023)
- [Quantile Regression for Distributional Reward Models in RLHF](https://mlsanity.com/paper/2409.10164/) (2024)
- [The Past, Present and Better Future of Feedback Learning in Large Language Models for Subjective Human Preferences and Values](https://mlsanity.com/paper/2310.07629/) (2023)

## Built on by (most cited first, 10 of 10)

- [What's In My Human Feedback? Learning Interpretable Descriptions of Preference Data](https://mlsanity.com/paper/2510.26202/) (2025)
- [Noise-corrected GRPO: From Noisy Rewards to Unbiased Gradients](https://mlsanity.com/paper/2510.18924/) (2025)
- [Value Drifts: Tracing Value Alignment During LLM Post-Training](https://mlsanity.com/paper/2510.26707/) (2025)
- [Automatically Finding Reward Model Biases](https://mlsanity.com/paper/2602.15222/) (2026)
- [EVALUESTEER: Measuring Reward Model Steerability Towards Values and Preferences](https://mlsanity.com/paper/2510.06370/) (2025)
- [Reward Models Inherit Value Biases from Pretraining](https://mlsanity.com/paper/2601.20838/) (2026)
- [One Bias After Another: Mechanistic Reward Shaping and Persistent Biases in Language Reward Models](https://mlsanity.com/paper/2603.03291/) (2026)
- [Unsupervised Elicitation of Moral Values from Language Models](https://mlsanity.com/paper/2601.17728/) (2026)
- [Discovering Implicit Large Language Model Alignment Objectives](https://mlsanity.com/paper/2602.15338/) (2026)
- [RewardExplainer: Learning Reward Model Explanations from Counterfactual Preference Feedback](https://mlsanity.com/paper/2609.33989/) (2026)

## Abstract (arXiv)

Reward modeling has emerged as a crucial component in aligning large language models with human values. Significant attention has focused on using reward models as a means for fine-tuning generative models. However, the reward models themselves -- which directly encode human value judgments by turning prompt-response pairs into scalar rewards -- remain relatively understudied. We present a novel approach to reward model interpretability through exhaustive analysis of their responses across their entire vocabulary space. By examining how different reward models score every possible single-token response to value-laden prompts, we uncover several striking findings: (i) substantial heterogeneity between models trained on similar objectives, (ii) systematic asymmetries in how models encode high- vs low-scoring tokens, (iii) significant sensitivity to prompt framing that mirrors human cognitive biases, and (iv) overvaluation of more frequent tokens. We demonstrate these effects across ten recent open-source reward models of varying parameter counts and architectures. Our results challenge assumptions about the interchangeability of reward models, as well as their suitability as proxies of complex and context-dependent human values. We find that these models can encode concerning biases toward certain identity groups, which may emerge as unintended consequences of harmlessness training -- distortions that risk propagating through the downstream large language models now deployed to millions.