About me
I am a fourth-year PhD student in CS at the University of Wisconsin-Madison, advised by Professor Frederic Sala and a member of the Sprocket Lab.
My research takes a data-centric approach to large language models (LLMs) and LLM agents, focusing on:
- evaluating them faithfully through verifiable outcomes, understanding skills, and metric design
- improving them through data selection, curation, and data synthesis
I also earned my M.S. in Computer Science (2026) and my B.S. (2023), both at the University of Wisconsin-Madison, the latter with a double major in Computer Science & Mathematics.
I am actively seeking internship opportunities in related fields. Please feel free to reach out to me at zhiqi [at] cs.wisc.edu, or take a look at my resume.
News
- September 2026 Fine-Tuning Small Reasoning Models for Quantum Field Theory accepted to MLST!
- July 2026 Two papers accepted to COLM 2026!
Publications
Conferences & Journals
Understanding and Dating LLM-Era Stylistic Changes via the Cohort Deviation Measure
Zhiqi Gao, Frederic Sala
Under review at NAACL 2027Models Can Model, But Can’t Bind: Structured Grounding in Text-to-Optimization
Zhiqi Gao*, Albert Ge*, Alexander Michael Berenbeim, Nathaniel D. Bastian, Frederic Sala
COLM (Conference on Language Modeling) 2026
Preliminary version (as “OR-LLM-Bench”) at ICLR 2026 DATA-FM Workshop
arXiv Code DatasetTest-Time Scaling Makes Overtraining Compute-Optimal
Nicholas Roberts, Sungjun Cho, Zhiqi Gao, Tzu-Heng Huang, Albert Wu, Gabriel Orlanski, Avi Trost, Kelly Buchanan, Aws Albarghouthi, Frederic Sala
COLM (Conference on Language Modeling) 2026
arXivFine-Tuning Small Reasoning Models for Quantum Field Theory
Nathaniel S. Woodward, Zhiqi Gao, Yurii Kvasiuk, Kendrick M. Smith, Frederic Sala, Moritz Münchmeyer
MLST (Machine Learning: Science and Technology) 2026
PAI (Conference on Physics and AI) 2026
arXiv Code DatasetPretrained Hybrids with MAD Skills
Nicholas Roberts, Samuel Guo, Zhiqi Gao, Satya Sai Srinath Namburi GNVV, Sonia Cromp, Chengjun Wu, Chengyu Duan, Frederic Sala.
COLM (Conference on Language Modeling) 2025
arXivTheoretical Physics Benchmark (TPBench) – a Dataset and Study of AI Reasoning Capabilities in Theoretical Physics
Daniel J.H. Chung, Zhiqi Gao, Yurii Kvasiuk, Tianyi Li, Moritz Münchmeyer, Maja Rudolph, Frederic Sala, Sai Chaitanya Tadepalli.
MLST (Machine Learning: Science and Technology) 2025
MMLS (Midwest Machine Learning Symposium) 2025 Lightning Talk
arXiv Website Dataset
Workshop Publications
Test-time Scaling Techniques in Theoretical Physics—A Comparison of Methods on the TPBench Dataset
Zhiqi Gao*, Tianyi Li*, Yurii Kvasiuk, Sai Chaitanya Tadepalli, Maja Rudolph, Daniel J.H. Chung, Frederic Sala, Moritz Münchmeyer
NeurIPS 2025 Machine Learning and the Physical Sciences (ML4PS) Workshop
arXivRe-Structuring CLIP’s Language Capabilities
Zhiqi Gao, Frederic Sala.
MMLS (Midwest Machine Learning Symposium) 2025
Poster PDF Blog Post
Industry Experience
Roblox Corporation
- Designed, developed, and deployed a full-stack Slack bot that combines a vector database with large language models (LLMs) to answer complex questions over internal knowledge via retrieval-augmented generation (RAG), outperforming the company’s existing Q&A Slack bot.
- Created an efficient data pipeline, ingesting diverse documents (Confluence, Stack Overflow, GitHub) and generating vector embeddings for rapid retrieval.
Teaching Experience
Comp Sci 540 — Introduction to Artificial Intelligence
Comp Sci 300 — Programming II
Service
Reviewer: NeurIPS 2024, 2025, 2026 · COLM 2026 · TMLR 2026
Undergraduate Projects
Tessellations on the Poincaré Half-Plane and Disk
Advisor: Professor Andrew Zimmer.
- NSF-funded REU in Analysis, UW-Madison, Summer 2022. Built an interactive tool for generating and exploring tessellations of the Poincaré disk and half-plane, used to teach hyperbolic geometry.
Poster PDF
Random Walks on Groups
Advisor: Dr. Nate Fisher
- Madison Experimental Mathematics Lab, Spring 2022. Mathematica simulations of random walks on ℤⁿ and the Heisenberg group, measuring expected travel distance, hitting times, and hitting-location distributions.
Poster PDF
Miscellaneous
Outside of research, I have spent many years studying classical Chinese literature and cultural history.
I have memorized around 500 works of classical Chinese poetry, ci lyric poetry, and prose, and have written over 80 original pieces in traditional forms, including poems, ci, prose, and couplets.
Before moving abroad, I served as a volunteer docent at Zhihua Temple (智化寺) in Beijing, and I am comfortable giving informal guided tours of Beijing’s historical sites, such as the Summer Palace and the Forbidden City.