Qirun Dai

Qirun Dai

CS PhD student at UChicago working on data and algorithms for LLM post-training.

About Me

I am a first-year CS PhD student at the University of Chicago, advised by Prof. Chenhao Tan. I am a member of the Chicago Human+AI (CHAI) lab, and affiliated with the broader Communication and Intelligence (C&I) group. I am also working closely with Prof. Hao Peng at the University of Illinois at Urbana-Champaign (UIUC).

Previously, I received my Bachelor’s degree in Artificial Intelligence from Fudan University in 2025. During my undergraduate study, I was fortunate to intern at UIUC with Prof. Hao Peng and Prof. Jiaqi W. Ma, and at Shanghai Jiao Tong University with Prof. Dequan Wang.

Research Interests

I am broadly interested in training data and algorithms for large language models (LLMs). My current research focuses on:

  • Extending the generalization scope of their reasoning and thinking behaviors. I believe that genuine thoughtful reasoning should be a robust behavior that can transfer to versatile domains (e.g., philosophical and social science writing) and formalisms (e.g., causal reasoning, agentic reasoning, continual learning, etc.), but the current math-centered post-training paradigm makes LLMs struggle towards this goal.

  • Evaluating and improving their complex, composite but underexplored real-world capabilities. I am especially interested in training LLMs to (1) proactively explore and discover, (2) leverage ambiguity in strategic communications, and (3) balance accurate reasoning with controllable creativity.

  • Data foundations throughout the lifecycle of language-centered AI. I am always fascinated by the role of data in shaping model behaviors, with a consistent interest in data curation, selection, attribution, and data-efficient supervision paradigms, especially in but not limited to the language modality.

Selected Publications

* denotes equal contributions, and † denotes equal advising.

  • Executable Counterfactuals: Improving LLMs’ Causal Reasoning Through Code
    Aniket Vashishtha*, Qirun Dai*, Hongyuan Mei, Amit Sharma†, Chenhao Tan†, Hao Peng†
    ICLR 2026; NeurIPS 2025 Workshop on FoRLM
    [paper] [code]
  • The Best Instruction-Tuning Data are Those That Fit
    Dylan Zhang, Qirun Dai, Hao Peng
    NeurIPS 2025 (Spotlight)
    [paper]
  • Improving Influence-based Instruction Tuning Data Selection for Balanced Learning of Diverse Capabilities
    Qirun Dai, Dylan Zhang, Jiaqi W. Ma, Hao Peng
    Findings of EMNLP 2025; ICLR 2025 Workshop on DATA-FM
    [paper]

News

[01/2026] One paper accepted by ICLR 2026! Check out how Executable Counterfactuals fills the gap in counterfactual reasoning evaluation by operationalizing abduction with verifiable supervision.

[12/2025] Attending NeurIPS 2025 and will present my work, GRAPE (Spotlight) and Executable Counterfactuals (FoRLM Workshop).

[09/2025] Officially started my CS PhD study as an honored member of the CHAI lab.

[09/2025] One paper accepted by EMNLP 2025 (Findings), and two papers by NeurIPS 2025 (Spotlight & Workshop)!

[06/2025] Officially graduated from Fudan University and received my Bachelor’s degree.

[04/2025] Attending ICLR 2025 and will present my work, Balanced and Influential Data Selection (BIDS), at the 2nd DATA-FM Workshop.