Zhen Wu

grad_profile.jpeg

zhenwu@cs.cmu.edu


Office: GHC 6717


LinkedIn / CV




Welcome to my website! :-)

I’m a PhD student at the Language Technologies Institute, School of Computer Science at Carnegie Mellon University. I’m fortunate to be advised by Professor Carolyn Rosé.

My research agenda is driven by a core question: How can we understand, guide, and evaluate model reasoning behaviors?

  1. Understanding how models encode information in the latent space.
  2. Analyzing failure modes to uncover why model behaviors diverge from expectations.
  3. Developing methods to steer model behaviors toward better alignment and transparency.
  4. Designing evaluation frameworks that provide fine-grained insights into model reasoning processes.

Previously, I obtained my B.S. degree in Computer Science and Mathematics with Honors from the University of Pittsburgh.

Fun fact about me: I was trained as a professional singer (similar to Bel canto). I also play flute, the cucurbit flute, and piano for fun. Sometimes I also read mysteries and watch classical concerts.

News

Sep 22, 2025 Ongoing work with high school students I mentor on mitigating LLM over-refusal with fine-grained refusal tokens has been accepted to the NeurIPS 2025 Mechanistic Interpretability Workshop (≈300 submissions)!
Aug 25, 2025 First day as a PhD student at CMU LTI!
Aug 08, 2025 Gave my MLT talk on Text-Graph representation complementarity [slides].
Mar 11, 2025 Paper about llm-powered multi-party collaboration infrastructure for supporting collaborative learning accepted to CSCL.
Jun 27, 2024 Invited talk about evaluating LLM social signal sensitivity @ Disney Research Group [slides]..
Jun 17, 2024 Paper about evaluating LLMs on sensitivity to language framing to elicit role-oriented social behaviors accepted to HuLLM workshop @ ACL.
May 16, 2024 Paper about leveraging machine-generated rationales to facilitate generalization in conversations accepted to ACL.
Aug 28, 2023 First day as an MLT student at CMU LTI!