安全对齐的LLMs在面对无害查询时，常常因表面词汇与恶意问题相似而产生不必要的拒绝，即"过度拒绝"问题。现有对齐方法要么只能减少过度拒绝，要么只能维持安全性，难以同时兼顾。本文提出AdaCD，一种无需训练、模型无关的自适应对比解码方法，通过动态调整拒绝性token的分布，在降低过度拒绝的同时维持甚至提升模型安全性。

Large Language Models (LLMs) need to be both helpful and safe, but achieving both is a major challenge. We propose MidPO, a Mixture of Experts (MoE) framework that fine-tunes two specialized experts for safety and helpfulness and uses a dynamic routing mechanism to adaptively balance them, outperforming existing methods.

Dual-perspective, dual-granularity fairness evaluation for LLM-based job recommendation: we assess bias from both the user and the recruiter sides, at individual and group levels.

AI-generated images are more visually stunning than ever, but how do they really stack up against natural, real-world photos? We introduce D-Judge, a large-scale benchmark designed to systematically investigate and quantify the discrepancies that remain.

After careful design and development, the official website of the Knowledge Intelligence Lab (KILab) at Sun Yat-sen University is now fully upgraded and online! The new website adopts modern design concepts, providing visitors with a clearer and more intuitive browsing experience.