Alibaba, Token Foundry · Researcher
About Me
Hello! I’m Yunzhi Yao (姚云志), a Researcher at Token Foundry, Alibaba, where I work on LLM interpretability and Agentic coding.
I received my Ph.D. degree in Computer Science and Technology from Zhejiang University in 2026, where I was fortunate to be advised by Prof. Huajun Chen and Prof. Ningyu Zhang. Before that, I earned my Bachelor’s degree in Software Engineering and Dual Degree in Finance from Shandong University in 2021.
Btw, I’m interested in Astrology. If you are interested in talking about research and life with me, please feel free to reach out!
Experience
Alibaba Tongyi Lab · Research Intern
Ant Group, Ling Team · Research Intern
University of California, Los Angeles (UCLA) · Visiting Research Scholar
Microsoft Research Asia (MSRA) · Research Intern
Research Interest
My research focuses on understanding and enhancing large language models (LLMs). I am currently working on LLM interpretability and agentic coding, and I am actively exploring the following topics:
- LLM Interpretability: Uncovering the internal mechanisms of LLMs — how knowledge is stored, how circuits emerge, and how we can steer model behavior in a principled way.
- Agentic Coding: Building and understanding LLM agents that write, edit, and reason about code to solve real-world software engineering tasks.
- Knowledge Mechanisms and Editing: Investigating how LLMs acquire, store, and utilize knowledge for reasoning, and developing concise and precise methods for model editing.
Selected Publications
Towards principled knowledge editing methods for large language model reasoning.
Nature Machine Intelligence
Reflection on Knowledge Editing: Charting the Next Steps.
CaKE: Circuit-aware Editing Enables Generalizable Knowledge Learners.
EMNLP 2025, ACL 2025 KnowFM, Oral @ NENLP 2025
Knowledge Circuits in Pretrained Transformers.
In Proceedings of the 38th Neural Information Processing Systems (NeurIPS 2024)
Editing Large Language Models: Problems, Methods, and Opportunities.
In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP 2023)
Knowledge Rumination for Pre-trained Language Models.
In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP 2023)
Schema-aware Reference as Prompt Improves Data-efficient Knowledge Graph Construction.
In Proceedings of the 46th International ACM SIGIR Conference (SIGIR 2023)
Kformer: Knowledge Injection in Transformer Feed-forward Layers.
In Proceedings of the 11th Natural Language Processing and Chinese Computing (NLPCC 2022)
Adapt-and-Distill: Developing Small, Fast and Effective Pretrained Language Models for Domains.
In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021 (ACL 2021)
Honors and Awards
- Excellent Graduate (2026), Zhejiang Province
- Excellent Graduate (2026), Zhejiang University
- Notable Reviewer (2025), ICLR
- Scholar Award (2024), NeurIPS
- Best Paper Award, Workshop LLM+KG @ VLDB 2024
- National Scholarship (2024), Zhejiang University
- Star of Tomorrow (2021), Microsoft Research Asia
- Excellent Graduate (2021), Shandong University
- National Scholarship (2018), Shandong University
Professional Services
- Area Chair: ACL, EMNLP, AAAI
- Reviewer: TKDE, ACL ARR, NeurIPS, ICML, ICLR, TASLP, COLM, AAAI, ACM MM, AISTATS, NLPCC
- Volunteer: AI TIME