8 papers
What Did the Agent Actually Do? Evidence-Grounded Oversight for Long-Horizon Agents
Zhongxiang Sun, Jiahao Yan, Hongkang Zhao +5
As agents take on long-horizon tasks, users shift from making individual decisions to overseeing autonomous execution. Yet the volume of agent activity and the fragmentation of sup…
Key Path Identification for Resolving Knowledge Conflicts via SAE-based Steering
Wenbo Zhang, Zhongxiang Sun, Zhiguang Han +1
Sparse autoencoder (SAE)-based steering has been widely used to address knowledge conflicts by guiding LLMs to be more faithful to the contextual knowledge. Existing methods usuall…
NeuroCogMap Reveals Cognitive Organization of Large Language Models
Zhongxiang Sun, Haolang Lu, Qiang Ma +11
Understanding how complex cognitive functions are organized within artificial systems is central to interpreting large language models (LLMs) and relating them to biological cognit…
Hidden States Know Where Reasoning Diverges: Credit Assignment via Span-Level Wasserstein Distance
Xinzhu Chen, Wei He, Huichuan Fan +7
Group Relative Policy Optimization (GRPO) performs coarse-grained credit assignment in reinforcement learning with verifiable rewards (RLVR) by assigning the same advantage to all…
Diagnosing Knowledge Conflict in Multimodal Long-Chain Reasoning
Jing Tang, Kun Wang, Haolang Lu +7
Multimodal large language models (MLLMs) in long chain-of-thought reasoning often fail when different knowledge sources provide conflicting signals. We formalize these failures und…
Omni-Safety under Cross-Modality Conflict: Vulnerabilities, Dynamics Mechanisms and Efficient Alignment
Kun Wang, Zherui Li, Zhenhong Zhou +8
Omni-modal Large Language Models (OLLMs) greatly expand LLMs' multimodal capabilities but also introduce cross-modal safety risks. However, a systematic understanding of vulnerabil…
Deep Search with Hierarchical Meta-Cognitive Monitoring Inspired by Cognitive Neuroscience
Zhongxiang Sun, Qipeng Wang, Weijie Yu +3
Deep search agents powered by large language models have demonstrated strong capabilities in multi-step retrieval, reasoning, and long-horizon task execution. However, their practi…
Streaming Hallucination Detection in Long Chain-of-Thought Reasoning
Haolang Lu, Minghui Pan, Ripeng Li +6
Long chain-of-thought (CoT) reasoning improves the performance of large language models, yet hallucinations in such settings often emerge subtly and propagate across reasoning step…