Liang Qiu

My research interests lie in Natural Language Processing and Conversational AI. My long-term goal is to enhance both the EQ—aligning AI systems with human values through mental state modeling and social reasoning—and the IQ—advancing their reasoning and decision-making capabilities through deep learning.

 

Currently, I focus on RL-based alignment algorithms that use agentic and reasoning critics to tackle scalable oversight challenges. My approach involves developing a multi-agent critic system capable of generating structured rubrics, executing relevant tools or reasoning steps, and dynamically dispatching sub-tasks based on context. This system provides reliable feedback signals to guide the LLM policy during RL post-training, improving its helpfulness, reasoning, and alignment with human intent.

 

I have served as a reviewer for ICML, NeurIPS, AAAI, ACL, EMNLP, SIGDIAL, NLPCC, IEEE/ACM Transactions on Audio, Speech, and Language Processing, and the Journal of Artificial Intelligence Research.

 

Outside of research, I enjoy a good latte and an occasional gin tonic.