Reward Model Interpretability via Optimal and Pessimal Tokens

TL;DR

This paper analyzes reward models for language models by examining their responses to all possible tokens, revealing biases, asymmetries, and sensitivities that impact their alignment with human values.

Contribution

It introduces a comprehensive method to interpret reward models through exhaustive token analysis, uncovering biases and heterogeneity in their scoring behaviors.

Findings

Reward models show significant heterogeneity even when trained on similar objectives.

Models exhibit asymmetries in scoring high- vs low-value tokens.

Reward models are sensitive to prompt framing, reflecting human biases.

Abstract

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.