About
Hi, I’m Shi Qiu (仇是), a senior undergraduate student at the School of Physics, Peking University. I’m also an incoming Ph.D in Computer Science at UNC Chapel-Hill, advised by Prof. Huaxiu Yao.
My research focuses on AI Agents and AI4Physics, with a particular emphasis on self-evolving autoresearch systems for scientific discovery — agents that iteratively generate hypotheses, design and run experiments, and write papers with minimal human intervention. My work explores two complementary directions: (1) building robust, self-evolving agentic frameworks that automate the full research loop; and (2) leveraging domain-specific physics knowledge to ground agents in rigorous scientific reasoning.
My open-source work has collectively earned over 22,000 GitHub stars, and I am a core contributor to AutoResearchClaw, MetaClaw, and Agents.
I’ve also been collaborating with Prof. Dawn Song’s team at UC Berkeley, and with the AI4Physics research group led by Professors Hua Xing Zhu, Qing-Hong Cao, and Ming-Xing Luo at the School of Physics, Peking University. Also, I’m honored to be the only undergrad advisor of AlphaXiv.
News
- Jul 2026I’ve launched a blog — a running collection of Agent Harness notes, kicking off with a deep dive on streaming reconstruction. Come take a look → JollySammy’s blog. Everyone’s welcome!
- May 2026Honored to be selected as a Beijing Municipal Outstanding Graduate, a Peking University Outstanding Graduate, and a Merit Student Pacesetter (三好学生标兵)!
- Mar 2026Our paper MetaClaw is out on arXiv — an agent that meta-learns and self-evolves just by talking to it!
- Mar 2026Our self-evolving autoresearch projects AutoResearchClaw and MetaClaw are released — together with our open-source ecosystem they have surpassed 22k+ GitHub stars!
- Jan 2026Our paper Paper2Figure was accepted to CVPR 2026!
- Jan 2026Our paper EduVisAgent was accepted to ICLR 2026! I’ll not come to Brazil — see you online @ICLR 2026!
- Nov 2025Honored to be the Advisor of AlphaXiv! Keep going for high-fidelity research!
Show earlier news
- Sep 2025Honored to receive the National Scholarship award!
- Sep 2025Our papers PHYBench, MJ-Video (Spotlight) and SuperGPQA were accepted by NeurIPS 2025! See you all in San Diego in December~
- Sep 2025Our paper GLIMPSE was accepted as an Oral at EMNLP 2025 Main Conference!
- May 2025We released PHYBench, the first large-scale reasoning benchmark within a physical context.
- Feb 2025Our paper MJ-Video was released on arXiv.
- Jan 2025Our paper MMIE was accepted by ICLR 2025 as an Oral presentation!
- 2024Honored to receive the Excellence in Research Award for contributions in AI and intelligent systems.
Contact
- Email: stephenshiqiu@gmail.com
- GitHub: StephenQSstarThomas
- Google Scholar: Shi Qiu
- X: @StephenQS0710
- Hugging Face: StarThomas1002
- Discord:
stephen_17949· WeChat:StephenQSqiuqiu· 小红书: 是是是一只吃播萨米 - Curriculum Vitae (PDF)
Publications
PHYBench: Holistic Evaluation of Physical Perception and Reasoning in Large Language Models
NeurIPS 2025 · Project
AutoResearchClaw: Self-Reinforcing Autonomous Research with Human-AI Collaboration
arXiv preprint, 2026 · Code
ClawArena: Benchmarking AI Agents in Evolving Information Environments
arXiv preprint, 2026
OmniMem: Autoresearch-Guided Discovery of Lifelong Multimodal Agent Memory
arXiv preprint, 2026
PRBench: End-to-End Paper Reproduction in Physics Research
arXiv preprint, 2026
MetaClaw: Just Talk — An Agent That Meta-Learns and Evolves in the Wild
arXiv preprint, 2026 · Code
An End-to-End Architecture for Collider Physics and Beyond
arXiv preprint, 2026 · Code
The Attack and Defense Landscape of Agentic AI: A Comprehensive Survey
arXiv preprint (survey), 2026
Show earlier publications
Iterated Agent for Symbolic Regression
arXiv preprint, 2025
GLIMPSE: Do Large Vision-Language Models Truly Think With Videos or Just Glimpse at Them?
EMNLP 2025 Main (Oral)
Minos: A Multimodal Evaluation Model for Bidirectional Generation Between Image and Text
arXiv preprint, 2025
From EduVisBench to EduVisAgent: A Benchmark and Multi-Agent Framework for Pedagogical Visualization
ICLR 2026
MJ-Video: Fine-Grained Benchmarking and Rewarding Video Preferences in Video Generation
NeurIPS 2025 (Spotlight)
MMIE: Massive Multimodal Interleaved Comprehension Benchmark for Large Vision-Language Models
ICLR 2025 (Oral) · Project
Agents: An Open-source Framework for Autonomous Language Agents
ICLR 2024 Workshop on LLM Agents · Code
Honors and Awards
- Beijing Municipal Outstanding Graduate (北京市优秀毕业生), 2026
- Peking University Outstanding Graduate (北京大学优秀毕业生), 2026
- Merit Student Pacesetter (三好学生标兵), 2026
- National Scholarship, 2025
- Excellence in Research Award, 2024
- Peking University Third Class Scholarship, 2024
- Shu Qi Scholarship, 2023
- Youth Award for Athletics, 2023
Education
Peking University — B.S. in Astronomy, School of Physics. Beijing, China, 2022 – Present. Overall GPA 3.64/4.0.
Research Experience
Sunblaze Group, UC Berkeley — Research Intern (2025 – present). Mentor: Prof. Dawn Song.
- Lead researcher of AgentXploit, an open-source, end-to-end agentic framework for automatic injection-point exploitation and validation.
AI4Physics Group, Peking University — Research Intern (2025 – present). Mentors: Prof. Hua Xing Zhu, Prof. Qing-Hong Cao, and Prof. Ming-Xing Luo.
- Lead researcher and initiator of PHYBench, coordinating 178 students from the School of Physics to build the first high-difficulty physics reasoning benchmark.
- Core contributor to IdeaSearch, developing LLM-assisted solutions for optimization problems in physics — HEP, astronomy, and fluid dynamics.
UNC-Chapel Hill — Research Intern (2024 – present). Mentor: Prof. Huaxiu Yao.
- Research on multimodal alignment and evaluation: co-led MMIE (a 20,103-sample interleaved comprehension benchmark with an MLLM-as-judge metric), proposed GLIMPSE (evaluating whether video LLMs truly watch videos), and released the MJ-Video reward model for video generation.
AIWaves — Research Intern (Jun – Aug 2023).
- Core developer and paper co-author of the open-source Agents framework for autonomous language agents (5,900+ stars on GitHub).
CUPT Team, Peking University (Mar – May 2023).
- Main Speaker; Third Prize in the school-wide China Undergraduate Physics Tournament.
Projects
- AutoResearchClaw (12.5k+ stars, 2025 – present) — a fully autonomous, self-evolving research system that turns an idea into a complete paper, chaining literature review, hypothesis generation, experiment design and execution, analysis, and writing.
- MetaClaw (3.4k+ stars, 2025 – present) — a meta-learning agent that continuously learns and self-evolves from natural-language interaction.
- ColliderAgent (2025 – present) — a decoupled, autonomous agentic architecture for high-energy physics phenomenology, automating the full collider analysis pipeline.
- Agents (5.9k+ stars, 2023 – present) — an open-source framework for data-centric, self-evolving autonomous language agents.
- CourseCommunity (261+ stars, 2024 – present) — an open-source, all-for-free platform providing university course notes and self-study guides.
Skills
- Research: AI agents, self-evolving autoresearch, multimodal LLMs, AI4Physics, LLM evaluation.
- Programming: Python, LaTeX, Markdown.
- Languages: native Mandarin; fluent English (TOEFL 112, CET-6 696).
Beyond Research
Athletics. President of the Peking University Jump-Rope Team — National Online Jump Rope Competition: ranked #29; Beijing College League: 1st place (Men’s 30s Double Under), 2nd place (Mixed 1-on-1 30s Single Under, broke the event record), 3rd place (Consecutive Triple Under). Member of the Physics Dept. Badminton Team — 3rd place in the Beijing University Cup team competition.
Leadership & community. Class Monitor of Class 8, 22nd Undergraduate Cohort; key member of the Peking University Blockchain Association.
Literature & writing. A personal collection of essays lives in my Literature repository. Favorite authors: Richard Powers, Amor Towles, and Anthony Doerr.
Photography. Collaborated with Mengshu Yu and Dannni Luo on a series of self-made postcards — see the Postcards collection.
Games. Clash Royale — ladder rank #31 and tournament rank #46 on the Chinese server; reached #1 using Evo RG (video proof); 3rd place in the Clash Royale Chinese University League. Battle of Balls — ranked #7 on the Zhejiang Province ladder, July 2018.
Trains & fishing. A passionate train enthusiast, and I enjoy fishing in my leisure time.