Bing Qin

This paper introduces the concept of Safety Geometry Collapse, demonstrating that multimodal inputs degrade the safety separation of LLMs, and proposes ReGap, a training-free method that adaptively corrects this drift to improve MLLM safety.

The paper argues that current search agents often verify existing knowledge rather than genuinely searching, and introduces LiveBrowseComp, a new benchmark to measure true evidence-driven discovery.

The paper proposes ESRT, an edge-cloud framework that achieves state-of-the-art, bandwidth-efficient, and privacy-preserving many-to-many speech translation across 45 languages by splitting the model inference.

DeepTool introduces a novel Process-Supervised Reinforcement Learning framework to enhance Tool-Integrated Reasoning by explicitly supervising and rewarding intermediate, interleaved deliberation steps during sequential tool use.

The paper introduces CultureForest, a new benchmark for evaluating Cultural Norm Grounded Reasoning in LLMs, demonstrating that models struggle to apply their cultural knowledge effectively in realistic, open-ended scenarios.