Configurable Reward Model for Balanced Safety Alignment
Zhengping Jiang · Mehran Khodabandeh · Akash Bharadwaj · Manik Bhandari · Mayur Srungarapu · Anqi Liu · Benjamin Van Durme · Li Chen
首次获取该候选人的完整档案及 StartLux 评估共 20 积分;每位用户单独解锁,之后查看免费。主动刷新 20 积分。
实时查找电子邮箱,费用不超过20积分;未找到邮箱不扣费。
Zhengping Jiang · Mehran Khodabandeh · Akash Bharadwaj · Manik Bhandari · Mayur Srungarapu · Anqi Liu · Benjamin Van Durme · Li Chen
Kai Hu · Abhinav Aggarwal · Mehran Khodabandeh · David Zhang · Eric Hsin · Li Chen · Ankit Jain · Matt Fredrikson · Akash Bharadwaj
This paper presents a novel Automated Red Teaming (ART) framework that shifts from example-based to policy-based evaluation, addressing critical limitations in scalability and validity. We define harmful content through abstract safety policies rather than specific static examples. We also introd...
Tinghao Xie · Yueqi Xie · Alireza Zareian · Shuming Hu · Felix Juefei-Xu · Xiaowen Lin · Ankit Jain · Prateek Mittal · Li Chen
Not Safe for Work (NSFW) image classifiers play a critical role in safeguarding text-to-image (T2I) systems. However, a concerning phenomenon has emerged in T2I systems – changes in text prompts that manipulate benign image elements can result in failed detection by NSFW classifiers – dubbed “ co...