About me
Hi! I am Xun Liu, a senior undergraduate majoring in Cybersecurity at University of Chinese Academy of Sciences, China. Before that, I developed my interest in Computer Science and Computational thinking in Olympiad in Informatics (OI).
My current research interest lies in Trustworthy Machine Learning, especially in Large Language Models. Surrounding this topic, my research explores jailbreaking, backdoor attacks and adversarial attacks.
Whatever my future endeavors are, I shall steadfastly view the integration of technology and humanities as my life's mission. In a world increasingly dominated by digital advancements, I believe that preserving the essence of human culture and values is paramount. My journey will be dedicated to harmonizing the precision and innovation of technology with the depth and insight of the humanities. Of course, this vision presents a long road ahead, not just for me personally, but also for the academic and industrial worlds.
Education
- B.S. in Cybersecurity
School of Cyber Security, University of Chinese Academy of Sciences
Aug. 2021-Present
Research Experience
- Secure Learning Lab, University of Illinois at Urbana-Champaign
Student Intern, advised by Prof. Bo Li.
Feb. 2024-Present - Institute of Computing Technology, Chinese Academy of Sciences
Research Assistant, advised by Prof. Fei Sun.
Sep. 2023-Present
Publications and Preprints
-
RedCode: Risky Code Execution and Generation Benchmark for Code Agents
Chengquan Guo*, Xun Liu*, Chulin Xie*, Andy Zhou, Yi Zeng, Zinan Lin, Dawn Song, Bo Li. NeurIPS 2024. - 🦋🌪 The Butterfly Effect of Model Editing: Few Edits Can Trigger Large Language Models Collapse
Wanli Yang, Fei Sun, Xinyu Ma, Xun Liu, Dawei Yin, Xueqi Cheng. ACL 2024.
Selected Projects
- Low-/Mid-Resource Language Jailbreaking LLM (python)
The Trojan Detection Challenge 2023 (LLM Edition) [Leaderboard]- Most Compute-Efficient Special Award of Large Model Subtrack of Red Teaming Track.
- Tools & technologies used: Python, PyTorch, LLaMa2.
- Select 87 low- and 24 mid-resource languages and translate the malicious prompts via Baidu/DeepL/Azure API. Use GPT-3.5 to supervise the tautological transformation of target prompts.
- Placed 5th in development phase of Large Model Subtrack of Red Teaming Track.
Teaching Experience
- Class 2311 (Cybersecurity Major)
Peer Mentor
University of Chinese Academy of Sciences - Introduction to Computer Science (Spring 2023)
Teaching Assistant
University of Chinese Academy of Sciences
Instructor: Prof. Zhiwei Xu
Honors and Awards
- First Class Academic Scholarship (中国科学院大学本科生一等学业奖学金, Top 5%), 2022
- National Scholarship Nomination (国家奖学金提名, Top 30/1600+), 2022
- Outstanding Student Cadre (中国科学院大学优秀学生干部), 2022,2023,2024
- Merit Student (中国科学院大学三好学生), 2022,2023,2024
- The 2021 ICPC Asia Nanjing Regional Contest, Bronze Medal, 2021
- National Olympiad in Informatics in Provinces (NOIp), First Prize (Senior Group) (提高组一等奖), 2018,2019
Paper Reading Presentations
Public Affairs
- President of the 10th Undergraduate Student Union of University of Chinese Academy of Sciences, 2023-2024
- President of the Pin Association (果壳良品社团社长), 2022-2023
- Cofounder and organizer of the No.2 High School Public Lecture (二中公益讲堂发起人、组织者), 2021