Minda Zhao outdoors on a sunny winter day

Minda Zhao

Student Researcher

Harvard University

About Me

I am a researcher at the Harvard AI and Robotics Lab at Harvard University, where I am fortunate to be advised by Mengyu Wang.

I received my S.M. in Health Data Science from Harvard University, where I had the privilege of being advised by Shilpa Nadimpalli Kobren and Isaac Kohane. Before Harvard, I received my B.S. in Data Science from Duke University and Duke Kunshan University.

My research focuses on understanding and improving the reliability and capabilities of generative foundation models, as well as advancing their real-world applications in specialized domains.

News

Publications

* Co-first authors.   † Corresponding authors.

Published & Accepted

  1. Evaluation pipeline: LLMs and probability distributions, batch and independent generation, and statistical diagnostics. ACL
    Large Language Models Are Bad Dice Players: LLMs Struggle to Generate Random Numbers from Statistical Distributions
    Minda Zhao, Yilun Du, and Mengyu Wang.
    64th Annual Meeting of the Association for Computational Linguistics (ACL), 2026.
    PDF Code
  2. TrailBlazer framework linking a history-aware reinforcement learning agent, prompt mutation, a helper LLM, and target-model feedback. COLM
    TrailBlazer: History-Guided Reinforcement Learning for Black-Box LLM Jailbreaking
    Sung-Hoon Yoon*, Ruizhi Qian*, Minda Zhao*, Weiyue Li*, and Mengyu Wang.
    Conference on Language Modeling (COLM), 2026.
    PDF Code
  3. Four-stage pipeline for LLM-as-a-judge: benchmark selection, inter-scale consistency, human–LLM agreement, and reliability analysis. npj AI
    Grading scale impact on LLM-as-a-judge: human-LLM alignment is highest on 0–5 grading scale
    Weiyue Li*, Minda Zhao*, Weixuan Dong*, Jiahui Cai*, Yuze Wei*, Michael Pocress, Yi Li, Wanyan Yuan, Xiaoyue Wang, Ruoyu Hou, Kaiyuan Lou, Wenqi Zeng, Yutong Yang, Yilun Du, and Mengyu Wang.
    npj Artificial Intelligence, 2026.
    PDF Code

Preprints

  1. RG-ICL study schematic showing medical imaging datasets, inference paradigms, frozen encoders, and representation-guided reference retrieval. arXiv
    Representation-guided in-context learning for medical image interpretation with multimodal large language models
    Minda Zhao*, Fangyu Hu*, Yan Luo*, Yutong Yang, Jiahui Cai, Kaichen Zhou, Manling Li, Paul Liang, Yilun Du, Lucy Q. Shen, and Mengyu Wang.
    arXiv preprint, 2026.
    PDF
  2. Five-phase workflow: rare-disease data curation, clinical vignette generation, validation, model evaluation, and analysis. arXiv
    Rare Diseases, Common Dilemmas: LLMs Prioritize Equal Resource Distribution over Patient Benefit in Decision-Making
    Minda Zhao†, Xu Han, Rishabh Goel, Maya Dagan, Noa Dagan, Adithya Madduri, Payal Chandak†, Shilpa Nadimpalli Kobren†, and Isaac S. Kohane†.
    arXiv preprint, 2026.
    PDF
  3. Emotional prompting pipeline connecting canonical and context-aligned emotional cues with LLM evaluation across benchmark tasks. arXiv
    Do Emotions in Prompts Matter? Effects of Emotional Framing on Large Language Models
    Minda Zhao, Yutong Yang, Chufei Peng, Rachel Gonsalves, Weiyue Li, Ruyi Yang, Zhixi Liu, and Mengyu Wang.
    arXiv preprint, 2026.
    PDF