Awesome RL with Human Feedback / 2023 / arXiv

The Trickle-down Impact of Reward (In-)consistency on RLHF

作者：Lingfeng Shen, Sihao Chen, Linfeng Song, Lifeng Jin, Baolin Peng, Haitao Mi, Daniel Khashabi, Dong Yu · 单位：Johns Hopkins University, University of Pennsylvania, Tencent AI Lab · 来源：arXiv · 发布日期：2023

奖励黑客/过优化奖励建模
新标签打开 AlphaXiv新标签打开 Code
The Trickle-down Impact of Reward (In-)consistency on RLHF figure
AlphaXiv 中文概览（可滚动查看）