Du Su

Reliable confidence is essential for trusting the outputs of LLMs, yet widely deployed post-trained LLMs (PoLLMs) typically compromise this trust with severe overconfidence. In contrast, we observe that their corresponding base LLMs often remain well-calibrated. This naturally motivates us to cal...

Safety alignment in Large Language Models (LLMs) inherently presents a multi-objective optimization conflict, often accompanied by an unintended degradation of general capabilities. Existing mitigation strategies typically rely on global gradient geometry to resolve these conflicts, yet they over...

Group Relative Policy Optimization (GRPO) has significantly advanced the reasoning ability of large language models (LLMs), particularly in their mathemat- ical reasoning performance. However, GRPO and related entropy regularization methods still struggle with token-level sparse-rewards, which is...

Investigating black-box jailbreak attacks is crucial for revealing the actual security risks faced by operational Large Language Models (LLMs). The primary challenge in black-box jailbreak attack is the absence of direct optimization signals, such as gradients, to guide the refinement of adversar...