Yuk-Kit Chan

Hello, I am a master's student in Computer Science and Engineering at the University of California, San Diego. My research focuses on evaluating and improving large language models and LLM agents, including factuality, tool use and benchmark reliability. I am currently supervised by Professor Jingbo Shang on how LLM agents gather and use evidence when exploring unfamiliar computational systems.

I earned my Bachelor of Engineering in Computer Science and Engineering from The Chinese University of Hong Kong, where I completed the ELITE stream, and graduated with First Class Honours and a minor in Mathematics. I was very fortunate to have been advised by Professor Wenxuan Wang.

Email  /  Github /  Google Scholar /  CV /  linkedin

profile photo
Education
University of California, San Diego, La Jolla, CA Sep 2026 - Jun 2028 (Expected)
Master of Science, Department of Computer Science and Engineering
Computer Science
The Chinese University of Hong Kong, Hong Kong Aug 2022 - Jul 2026
Bachelor of Engineering, Department of Computer Science and Engineering
Artificial Intelligence – Systems & Technologies, First Class Honours, ELITE Stream, Minor in Mathematics
Publications
Position: Code Benchmarks Should Prioritize Rigor, Reliability, and Reproducibility.
Jialun Cao, Yuk-Kit Chan*, Zixuan Ling*, Wenxuan Wang, Shuqing Li,
Mingwei Liu, Ruixi Qiao, Yuting Han, Chaozheng Wang, Boxi Yu, Pinjia He, Shuai Wang,
Zibin Zheng, Michael R. Lyu, Shing-Chi Cheung
ICML, 2026
arXiv | ICML

We conducted a decade-scale survey of 672 code benchmarks, found a growing awareness gap in rigorous benchmark practice, and respond by proposing HOW2BENCH

Identifying the Achilles' Heel: An Iterative Method for Dynamically Uncovering Factual Errors in Large Language Models
Wenxuan Wang, Yuk-Kit Chan*, Zixuan Ling*, Juluan Shi*,
Youliang Yuan, Jen-tse Huang, Yifei Zhang, Wenxiang Jiao, Zhaopeng Tu, Michael R. Lyu
Findings of ACL, 2026
arXiv | ACL | Github

We introduce HalluHunter, a fully automated knowledge-graph framework that generates single- and multi-hop factuality questions and adaptively targets LLM weaknesses

Learning to Ask: When LLM Agents Meet Unclear Instruction.
Wenxuan Wang, Juluan Shi*, Zixuan Ling*, Yuk-Kit Chan*, Chaozheng Wang,
Cheryl Lee, Youliang Yuan, Jen-tse Huang, Wenxiang Jiao, and Michael R. Lyu
EMNLP, 2025
arXiv | EMNLP | Github

We study tool-use under imperfect user instructions by analyzing real queries and introducing Noisy ToolBench, showing LLMs often hallucinate missed arguments; we then propose Ask-when-Needed (AwN) and an automated ToolEvaluator that improve tool-use accuracy over existing methods.

Current Research & Projects
Market Lab: A Research Workbench for Testing AI Trading Agents
Sole developer
Open source, v1.0.0, 2026
Website | Github | PyPI

A local platform for building LLM trading agents and testing them against human, rule-based and neural-network baselines on historical market data, with a C++ market engine, replayable records and statistical evaluation.





Last updated Nov 2026.
Website template from Jon Barron.