Hi, I’m Min Cai (蔡旻). I’m an incoming PhD student at the University of Alberta, supervised by Dr. Xi Ye. Previously, I was an M.S. student graduated from Shenzhen University, where I was supervised by Prof. Haodi Zhang. Before that, I obtained my B.A. in Translation from Beijing Language and Culture University. Currently, I’m interning at Zhipu AI, mentored by Dr. Dan Zhang. I also work closely with Dr. Ziniu Hu, Dr. Shichang Zhang, and Dr. Difan Zou.

I have broad interests in ML and NLP, particularly in understanding the mechanisms behind neural language models, developing LLM agents capable of solving complex problems, and enhancing LLM reasoning. My primary focus is on inference-time algorithms for alignment and reasoning in LLMs.

Research areas:

  • Interpreting and controlling LLM behaviors — mechanistic understanding for better alignment with human values (e.g., SelfControl)
  • LLM Agents — solving complex tasks via multi-agent games and strategic planning (e.g., AvalonBench)
  • LLM Reasoning — inference-time algorithms such as Monte Carlo tree search and representation engineering (e.g., Strategist)

Selected Publications

Paper figure
Preprint ★ Outstanding Paper — New England NLP Workshop

How Post-Training Reshapes LLMs: A Mechanistic View on Knowledge, Truthfulness, Refusal, and Confidence

Hongzhe Du*, Weikai Li*, Min Cai, Karim Saraipour, Zimin Zhang, Himabindu Lakkaraju, Yizhou Sun, Shichang Zhang (*equal contribution)

Studied how post-training reshapes LLMs on knowledge storage, truthfulness, refusal and confidence, using causal tracing, linear probing and entropy neurons.

Strategist figure
ICLR 2025

Strategist: Learning Strategic Skills by LLMs via Bi-Level Tree Search

Jonathan Light, Min Cai, Weiqin Chen, Guanzhi Wang, Xiusi Chen, Wei Cheng, Yisong Yue, Ziniu Hu

International Conference on Learning Representations (ICLR 2025) · Covered by State of AI Report 2024

Strategist uses LLMs to acquire new strategic skills for multi-agent games through a bi-level tree search self-improvement process.

SelfControl figure
ICML 2024 Workshop

Self-Control of LLM Behaviors by Compressing Suffix Gradient into Prefix Controller

Min Cai, Yuchen Zhang, Shichang Zhang, Fan Yin, Dan Zhang, Difan Zou, Yisong Yue, Ziniu Hu

Foundation Models in the Wild & Mechanistic Interpretability Workshop, ICML 2024

An inference-time LLM control method that leverages self-evaluation to steer model behaviors through representation engineering.

AvalonBench figure
NeurIPS 2023 Workshop

AvalonBench: Evaluating LLMs Playing the Game of Avalon

Jonathan Light*, Min Cai*, Sheng Shen, Ziniu Hu (*equal contribution)

Foundation Models for Decision Making Workshop, NeurIPS 2023

A benchmark exploring the potential of LLM agents in Resistance Avalon, a strategic social deduction game requiring reasoning and coalition building.

Beyond Academics

In my spare time, I enjoy playing and listening to music — jazz, classical, R&B, and more. I also play the Pokémon Trading Card Game (PTCG) and love exploring new foods.