2 papers
cs.LG2026
Reproducing, Analyzing, and Detecting Reward Hacking in Rubric-Based Reinforcement Learning
Xuekang Wang, Zhuoyuan Hao, Shuo Hou +3
Rubric-based reinforcement learning (RL) uses an LLM-as-a-Judge (LaaJ) to score model outputs according to rubrics as rewards. However, policy models may exploit latent biases in t…
cs.LG2025
Speculative Safety-Aware Decoding
Xuekang Wang, Shengyu Zhu, Xueqi Cheng
Despite extensive efforts to align Large Language Models (LLMs) with human values and safety rules, jailbreak attacks that exploit certain vulnerabilities continuously emerge, high…