Wenxi Dai 戴文希
Research Assistant · The Chinese University of Hong Kong, Shenzhen
I'm currently working with Prof. Mengnan Du as a research assistant at the Chinese University of Hong Kong, Shenzhen. I recently completed my master's degree in Computer Science at Rutgers, New Brunswick, where I conducted research under the supervision of Prof. Dimitris N. Metaxas.
My research broadly focuses on LLMs, including interpretability, post-training, and agents. More specifically, I have recently been interested in:
- Understanding how training data and optimization shape models' internal representations and behaviors, and leveraging these insights to develop more efficient post-training methods.
- Investigating the internal mechanisms underlying model capabilities, and exploring how these insights can enable more precise control over model and agent behaviors.
wd232 [at] scarletmail [dot] rutgers [dot] edu
Research Interests
I'm open to collaborations on related topics. Feel free to drop me an email.
News
- Oct 2026New preprint on the mechanisms of membership signals in LLMs: From Representation to Readout.
- Oct 2026New preprint on decision-making under uncertainty for tool-using agents: Agents' Last Doubt.
- Nov 2025Preprint released: Token-Controlled Re-ranking for Sequential Recommendation via LLMs.
- 2025Serving as a reviewer for AAAI 2026.
Publications
* denotes equal contribution
Education
Rutgers University, New Brunswick
M.S. in Computer Science
Jilin University
Bachelor's degree, College of Computer Science and Technology
Professional Services
- Reviewer, AAAI 2026
