10 papers
LLM-VA: Resolving the Jailbreak-Overrefusal Trade-off via Vector Alignment
Haonan Zhang, Dongxia Wang, Yi Liu +2
Safety-aligned LLMs suffer from two failure modes: jailbreak (answering harmful inputs) and over-refusal (declining benign queries). Existing vector steering methods adjust the mag…
MiroFlow: Towards High-Performance and Robust Open-Source Agent Framework for General Deep Research Tasks
Shiqian Su, Sen Xing, Xuan Dong +13
Despite the remarkable progress of large language models (LLMs), the capabilities of standalone LLMs have begun to plateau when tackling real-world, complex tasks that require inte…
ORFuzz: Fuzzing the "Other Side" of LLM Safety -- Testing Over-Refusal
Haonan Zhang, Dongxia Wang, Yi Liu +5
Large Language Models (LLMs) increasingly exhibit over-refusal - erroneously rejecting benign queries due to overly conservative safety measures - a critical functional flaw that u…
When Harmful Content Gets Camouflaged: Unveiling Perception Failure of LVLMs with CamHarmTI
Yanhui Li, Qi Zhou, Zhihong Xu +3
Large vision-language models (LVLMs) are increasingly used for tasks where detecting multimodal harmful content is crucial, such as online content moderation. However, real-world h…
CrossPL: Evaluating Large Language Models on Cross Programming Language Code Generation
Zhanhang Xiong, Dongxia Wang, Yuekang Li +2
As large language models (LLMs) become increasingly embedded in software engineering workflows, a critical capability remains underexplored: generating correct code that enables cr…
Sticking to the Mean: Detecting Sticky Tokens in Text Embedding Models
Kexin Chen, Dongxia Wang, Yi Liu +2
Despite the widespread use of Transformer-based text embedding models in NLP tasks, surprising 'sticky tokens' can undermine the reliability of embeddings. These tokens, when repea…