Reward Model Interpretability via Optimal and Pessimal Tokens
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2025
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866917242714718208 |
|---|---|
| author | Christian, Brian Kirk, Hannah Rose Thompson, Jessica A. F. Summerfield, Christopher Dumbalska, Tsvetomira |
| author_facet | Christian, Brian Kirk, Hannah Rose Thompson, Jessica A. F. Summerfield, Christopher Dumbalska, Tsvetomira |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_07326 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Reward Model Interpretability via Optimal and Pessimal Tokens Christian, Brian Kirk, Hannah Rose Thompson, Jessica A. F. Summerfield, Christopher Dumbalska, Tsvetomira Computation and Language Artificial Intelligence Computers and Society Machine Learning I.2.6; I.2.7; H.5.2; J.4; K.4.2 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. |
| title | Reward Model Interpretability via Optimal and Pessimal Tokens |
| topic | Computation and Language Artificial Intelligence Computers and Society Machine Learning I.2.6; I.2.7; H.5.2; J.4; K.4.2 |
| url | https://arxiv.org/abs/2506.07326 |