Hugging FaceKeyu Li
About Me
I am Keyu Li (李克钰), a second-year Ph.D. student at Shanghai Jiao Tong University and the Shanghai Innovation Institute. My research interests focus on Multi-Agent Systems, LLM post-training, and frontier AI evaluation.
I am currently collaborating with Prof. Dequan Wang and Prof. Pengfei Liu.
You can download my latest resumes here:
Recent News
- March 2026 — Excited to announce that two papers are accepted by ICLR 2026:
- January 2026 — Excited to share AgencyBench: Benchmarking the Frontiers of Autonomous Agents in 1M-Token Real-World Contexts (paper).
- September 2025 — Excited to release Limi: Less is More for Agency (paper).
- August 2025 — Excited to release DatasetResearch: Benchmarking Agent Systems for Demand-Driven Dataset Discovery (paper).
Education
- Ph.D. in Computer Science and Technology, Shanghai Jiao Tong University (2024–2029, expected)
- Research topics: AI agents, multi-agent systems, and LLM post-training
- Advisors: Prof. Dequan Wang and Prof. Pengfei Liu
- B.S. in Mathematics and Applied Mathematics, Shanghai Jiao Tong University (2020–2024)
- Academic score: 92.6/100, GPA: 4.03/4.3, Rank: 4/45
- Honors: Shanghai Outstanding Graduate, Guo Chenchen Scholarship (SJTU), Undergraduate Scholarship, Excellent League Member, Three-Good Student Award
- Bachelor thesis: Towards Visualizing and Understanding Bayesian Flow Networks
- Qingdao No.2 High School, Shandong (2017–2020)
- Honors: Outstanding Graduate, Excellent League Member, Three-Good Student Award
Personal Interests
- Deep user of OpenAI Codex and Claude Code for research and engineering workflows
- Long-term investor focused on high-quality companies and long-horizon value creation
- Enthusiast of basketball, football, and fitness training
Publications
Selected works (from my latest English CV):
- AgencyBench: Benchmarking the Frontiers of Autonomous Agents in 1M-Token Real-World Contexts (2026) — arXiv
- DaVinci-Agency: Unlocking Long-Horizon Agency Data-Efficiently (2026) — arXiv
- Interaction as Intelligence Part II: Asynchronous Human-Agent Rollout for Long-Horizon Task Training (2025) — arXiv
- DatasetResearch: Benchmarking Agent Systems for Demand-Driven Dataset Discovery (2025) — arXiv
- InnovatorBench: Evaluating Agents’ Ability to Conduct Innovative AI Research (2025) — arXiv
- Limi: Less is More for Agency (2025) — arXiv
- KAN-Mixer: Kolmogorov-Arnold Networks for Gene Expression Prediction in Plant Species (ECCV 2024) — Springer