Summary of Beyond Scalar Reward Model: Learning Generative Judge From Preference Data, by Ziyi Ye et al.
Beyond Scalar Reward Model: Learning Generative Judge from Preference Data
by Ziyi Ye, Xiangsheng Li, Qiuchi Li, Qingyao Ai, Yujia Zhou, Wei Shen, Dong Yan, Yiqun Liu
First submitted to arxiv on: 1 Oct 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A novel approach to learning from preference feedback aligns large language models with human values by leveraging their generation capabilities. The conventional method uses a scalar reward model that connects a value head with the LLM, but this lacks interpretability and is prone to biases. This paper proposes Direct Preference Optimization (DPO) using self-generated contrastive judgment pairs with natural language rationales, which ensures interpretability and robustness against bias without an additional reward head. The proposed generative judge, Con-J, achieves comparable performance to the scalar model while demonstrating superior interpretability and robustness in encoding human preferences. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models can be trained to align with human values by using preference feedback. Normally, this is done with a scalar reward model that connects a value head with the LLM, but this can be biased and hard to understand. This paper shows how to do it differently. Instead of using a scalar model, they use a special kind of judge that generates reasons for its judgments. This makes the results more understandable and less likely to be influenced by biases in the data. The new approach works just as well as the old one and is better in some ways. |
Keywords
* Artificial intelligence * Optimization
