Automatically Finding Reward Model Biases
This work addresses the issue of spurious rewards in LLM post-training for researchers and practitioners, though it is incremental as it builds on existing concerns about reward model biases.
The paper tackles the problem of automatically identifying biases in reward models for large language models, such as favoring redundant spacing or hallucinated content, and demonstrates that their method can recover known biases and discover new ones, with evolutionary iteration outperforming flat best-of-N search.
Reward models are central to large language model (LLM) post-training. However, past work has shown that they can reward spurious or undesirable attributes such as length, format, hallucinations, and sycophancy. In this work, we introduce and study the research problem of automatically finding reward model biases in natural language. We offer a simple approach of using an LLM to iteratively propose and refine candidate biases. Our method can recover known biases and surface novel ones: for example, we found that Skywork-V2-8B, a leading open-weight reward model, often mistakenly favors responses with redundant spacing and responses with hallucinated content. In addition, we show evidence that evolutionary iteration outperforms flat best-of-N search, and we validate the recall of our pipeline using synthetically injected biases. We hope our work contributes to further research on improving RMs through automated interpretability methods.