Lang Gao 高朗

Lang Gao, Jinghui Zhang, Wei Liu, Fengxian Ji, Chenxi Wang, Zirui Song, Akash Ghosh, Youssef Mohamed, Preslav Nakov, and Xiuying Chen

"By relaxing the orthogonality assumption in the Linear Representation Hypothesis, we introduce a cylindrical structure that explains why steering directions often interact and lead to unstable behaviors."

Paper Code Dataset Talk · Cohere.ai

Lang Gao, Xuhui Li, Chenxi Wang, Mingzhe Li, Wei Liu, Zirui Song, Jinghui Zhang, Rui Yan, Preslav Nakov, and Xiuying Chen

"What if I told you your AI detector can still achieve a high AUC on random tokens? The first work to reveal the weak transferability of machine-generated text detectors on personalized content, along with its mechanistic interpretation."

Lang Gao, Kaiyang Wan, Wei Liu, Chenxi Wang, Zirui Song, Zixiang Xu, Yanbo Wang, Veselin Stoyanov, and Xiuying Chen

"BiasLens is a new interpretable method that directly examines concept representations inside LLMs to detect hidden biases, without relying on any human-labeled data."

Lang Gao, Jiahui Geng, Xiangliang Zhang, Preslav Nakov, and Xiuying Chen

"We interpret the common mechanisms of diverse LLM jailbreak attacks in the activation space and propose an efficient defense method."

Kaiyang Wan, Lang Gao, Honglin Mu, Preslav Nakov, Yuxia Wang, Xiuying Chen

"An LLM's accuracy suffers a severe breakdown once the required information exceeds its internal capacity in complex multi-hop reasoning scenarios."

Wei Liu, Zhongyu Niu, Lang Gao, Zhiying Deng, Jun Wang, Haozhao Wang, and Ruixuan Li

"An interpretable, causal learning paradigm that simultaneously avoids spurious correlations in data and traditional self-interpretable models."

Yunfei Xie*, Ce Zhou*, Lang Gao*, Juncheng Wu*, Xianhang Li, Hong-Yu Zhou, Sheng Liu, Lei Xing, James Zou, Cihang Xie, and Yuyin Zhou (*: Joint first authors)

"A comprehensive, large-scale multimodal dataset for medical vision-language models."

Yanbo Wang, Zixiang Xu, Yue Huang, Xiangqi Wang, Zirui Song, Lang Gao, Chenxi Wang, Xiangru Tang, Yue Zhao, Arman Cohan, Xiangliang Zhang, and Xiuying Chen

"DyFlow is a dynamic workflow framework for LLM-based agents that adapts its reasoning steps in real time using intermediate feedback, enabling better generalization across diverse tasks."