# Interpreting Language Reward Models via Contrastive Explanations

arXiv: https://arxiv.org/abs/2411.16502 · HTML: https://mlsanity.com/paper/2411.16502/

## Key facts

- Cited by 9 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 47 references that are in the map.
- First posted on arXiv 25 Nov 2024.

## 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)

- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks](https://mlsanity.com/paper/1908.10084/) (2019)
- ["Why Should I Trust You?": Explaining the Predictions of Any Classifier](https://mlsanity.com/paper/1602.04938/) (2016)
- [Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2204.05862/) (2022)
- [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://mlsanity.com/paper/2006.03654/) (2020)
- [Explanation in Artificial Intelligence: Insights from the Social Sciences](https://mlsanity.com/paper/1706.07269/) (2017)
- [Scaling Laws for Reward Model Overoptimization](https://mlsanity.com/paper/2210.10760/) (2022)
- [Counterfactual Explanations without Opening the Black Box: Automated Decisions and the GDPR](https://mlsanity.com/paper/1711.00399/) (2017)
- [OpenAssistant Conversations -- Democratizing Large Language Model Alignment](https://mlsanity.com/paper/2304.07327/) (2023)
- [RewardBench: Evaluating Reward Models for Language Modeling](https://mlsanity.com/paper/2403.13787/) (2024)
- [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)
- [Evaluating Large Language Models at Evaluating Instruction Following](https://mlsanity.com/paper/2310.07641/) (2023)
- [A Long Way to Go: Investigating Length Correlations in RLHF](https://mlsanity.com/paper/2310.03716/) (2023)
- [Is DPO Superior to PPO for LLM Alignment? A Comprehensive Study](https://mlsanity.com/paper/2404.10719/) (2024)
- [HelpSteer2: Open-source dataset for training top-performing reward models](https://mlsanity.com/paper/2406.08673/) (2024)
- [Polyjuice: Generating Counterfactuals for Explaining, Evaluating, and Improving Models](https://mlsanity.com/paper/2101.00288/) (2021)
- [HelpSteer: Multi-attribute Helpfulness Dataset for SteerLM](https://mlsanity.com/paper/2311.09528/) (2023)
- [SteerLM: Attribute Conditioned SFT as an (User-Steerable) Alternative to RLHF](https://mlsanity.com/paper/2310.05344/) (2023)
- [OffsetBias: Leveraging Debiased Data for Tuning Evaluators](https://mlsanity.com/paper/2407.06551/) (2024)
- [RLHF Deciphered: A Critical Analysis of Reinforcement Learning from Human Feedback for LLMs](https://mlsanity.com/paper/2404.08555/) (2024)
- [Supporting Qualitative Analysis with Large Language Models: Combining Codebook with GPT-3 for Deductive Coding](https://mlsanity.com/paper/2304.10548/) (2023)
- [Dense Reward for Free in Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2402.00782/) (2024)
- [LLM-in-the-loop: Leveraging Large Language Model for Thematic Analysis](https://mlsanity.com/paper/2310.15100/) (2023)
- [LLM-Assisted Content Analysis: Using Large Language Models to Support Deductive Coding](https://mlsanity.com/paper/2306.14924/) (2023)
- [On Generating Plausible Counterfactual and Semi-Factual Explanations for Deep Learning](https://mlsanity.com/paper/2009.06399/) (2020)
- [Faithful Explanations of Black-box NLP Models Using LLM-generated Counterfactuals](https://mlsanity.com/paper/2310.00603/) (2023)
- [Generating Plausible Counterfactual Explanations for Deep Transformers in Financial Text Classification](https://mlsanity.com/paper/2010.12512/) (2020)
- [LLMs for Generating and Evaluating Counterfactuals: A Comprehensive Study](https://mlsanity.com/paper/2405.00722/) (2024)

## Built on by (most cited first, 9 of 9)

- [Representation Consistency for Accurate and Coherent LLM Answer Aggregation](https://mlsanity.com/paper/2506.21590/) (2025)
- [Automatically Finding Reward Model Biases](https://mlsanity.com/paper/2602.15222/) (2026)
- [Multi-Domain Explainability of Preferences](https://mlsanity.com/paper/2505.20088/) (2025)
- [Teach a Reward Model to Correct Itself: Reward Guided Adversarial Failure Discovery for Robust Reward Modeling](https://mlsanity.com/paper/2507.06419/) (2025)
- [Argumentative Ensembling for Robust Recourse under Model Multiplicity](https://mlsanity.com/paper/2506.20260/) (2025)
- [Discovering Implicit Large Language Model Alignment Objectives](https://mlsanity.com/paper/2602.15338/) (2026)
- [Informative Semi-Factuals for XAI: The Elaborated Explanations that People Prefer](https://mlsanity.com/paper/2603.17534/) (2026)
- [Understanding helpfulness and harmless tension in reward models](https://mlsanity.com/paper/2606.13209/) (2026)
- [RewardExplainer: Learning Reward Model Explanations from Counterfactual Preference Feedback](https://mlsanity.com/paper/2609.33989/) (2026)

## Abstract (arXiv)

Reward models (RMs) are a crucial component in the alignment of large language models' (LLMs) outputs with human values. RMs approximate human preferences over possible LLM responses to the same prompt by predicting and comparing reward scores. However, as they are typically modified versions of LLMs with scalar output heads, RMs are large black boxes whose predictions are not explainable. More transparent RMs would enable improved trust in the alignment of LLMs. In this work, we propose to use contrastive explanations to explain any binary response comparison made by an RM. Specifically, we generate a diverse set of new comparisons similar to the original one to characterise the RM's local behaviour. The perturbed responses forming the new comparisons are generated to explicitly modify manually specified high-level evaluation attributes, on which analyses of RM behaviour are grounded. In quantitative experiments, we validate the effectiveness of our method for finding high-quality contrastive explanations. We then showcase the qualitative usefulness of our method for investigating global sensitivity of RMs to each evaluation attribute, and demonstrate how representative examples can be automatically extracted to explain and compare behaviours of different RMs. We see our method as a flexible framework for RM explanation, providing a basis for more interpretable and trustworthy LLM alignment.