Kaijian (Kai) Zou

avatar.jpg

Hello, my name is Kaijian Zou. My Chinese name is 邹凯键. People usually call me Kai.

I am a 4th-year Ph.D. Candidate at the University of Michigan, advised by Prof. Lu Wang and a member of the LAUNCH Group.

I earned my B.S. in Computer Science from Cornell University, where I had the privilege of working with Prof. Claire Cardie on summarization research.

My research interests lie in Natural Language Processing, Deep Learning, and Machine Learning. I have worked on the code reasoning of LLMs and the evaluation of long-context language models. Currently, I am working on improving SWE agents with the step-level verifier and investigating long-horizon agentic RL.

My email is zkj[last_name]@umich.edu.

news

Oct 22, 2026 SWE-VERIFIER: Selective Verification for Cost-Efficient Coding Agents is out on arXiv. We introduce a lightweight model that selectively triggers candidate sampling and verification to improve coding-agent performance with low additional cost. Read our tweet here.
Oct 20, 2026 Predicting Privileged Information Effectiveness for On-Policy Self-Distillation is out on arXiv. We investigate which privileged information improves agentic post-training and how to assess it before training. Read our tweet here.
Jun 16, 2026 The Wrong Kind of Right: Quantifying and Localizing Misfired Alignment in LLMs is out on arXiv.
May 26, 2026 Excited to start my summer internship at Scale AI as an ML researcher, focusing on agentic post-training.
Apr 30, 2026 LiveOIBench: Can Large Language Models Outperform Human Contestants in Informatics Olympiads? accepted at ICML 2026. See you in Seoul.
Aug 20, 2025 SYNC: A Synthetic Long-Context Understanding Benchmark for Controlled Comparisons of Model Capabilities accepted at EMNLP 2025.
May 19, 2025 Excited to start my internship at Cresta as a ML researcher.
May 15, 2025 On Many-Shot In-Context Learning for Long-Context Evaluation accepted at ACL 2025. See you at Vienna.
Nov 11, 2024 Retrieval or Global Context Understanding? On Many-Shot In-Context Learning for Long-Context Evaluation is out on arXiv.
Oct 09, 2024 Passed my prelim exam!
Oct 07, 2023 Two papers accpeted at EMNLP 2023! Crossing the Aisle: Unveiling Partisan and Counter-Partisan Events in News Reporting and All Things Considered: Detecting Partisan Events from News Media with Cross-Article Comparison
Aug 29, 2022 Started my Phd at University of Michigan :smile:

selected publications

  1. EMNLP-Findings
    Crossing the Aisle: Unveiling Partisan and Counter-Partisan Events in News Reporting
    Kaijian Zou, Xinliang Frederick Zhang, Winston Wu, and 2 more authors
    2023
    PDF Data Poster Slides
  2. ACL
    On Many-Shot In-Context Learning for Long-Context Evaluation
    Kaijian Zou, Muhammad Khalifa, and Lu Wang
    2025
    PDF Code Data Poster Slides
  3. ICML
    LiveOIBench: Can Large Language Models Outperform Human Contestants in Informatics Olympiads?
    Kaijian Zou, Aaron Xiong, Yunxiang Zhang, and 6 more authors
    2026
    DOI arXiv HTML PDF Code Data Website
  4. arXiv
    Predicting Privileged Information Effectiveness for On-Policy Self-Distillation
    Kaijian Zou, Emiliano Penaloza, MohammadHossein Rezaei, and 5 more authors
    arXiv preprint arXiv:2610.22512, 2026
    Paper Code Data Tweet
  5. arXiv
    SWE-VERIFIER: Selective Verification for Cost-Efficient Coding Agents
    Kaijian Zou, Vaibhav Chennareddy, Yunxiang Zhang, and 5 more authors
    arXiv preprint arXiv:2610.30284, 2026
    Paper