Shuzheng Si

Ph.D. student, Tsinghua University

Hi, I’m Shuzheng Si, a second-year CS Ph.D. Candidate at Tsinghua University. I am lucky to be advised by Prof. Maosong Sun from the TsinghuaNLP Lab. Before this, I obtained my master’s degree from Peking University, where I was supervised by Prof. Baobao Chang at the Institute of Computational Linguistics. My research interests lie in Natural Language Processing and Large Language Models. Recently, I have been interested in a fundamental challenge for LLMs: Hallucinations.

• 📖 Hallucination Attribution — Understanding Hallucinations at the Source: My first line of research aims to explore and understand how training data contributes to hallucinations (NOVA, GATEAU, NUGGETS). This line of research mainly focuses on combining both qualitative and quantitative analyses of training data, guiding data collection to reduce LLM hallucinations.
• 🔧 Hallucination Mitigation — Reducing Hallucinations in LLMs: This line of research focuses on designing effective strategies to reduce hallucinations, covering text generation (CANOE, LingoEDU), multi-modal scenarios (MMICL, LACING), and agentic tasks (SpokenWOZ, EAGLET, RhinoInsight). These methods provide practical solutions to reduce hallucinations in real-world applications.
• 🔎 Hallucination Detection — Identifying Hallucinations in the Wild: This research attempts to identify hallucinated responses generated by LLMs and prevent such hallucinated responses from being provided to users (FaithLens, Infi-Check). By enabling the timely hallucination detection and practical interventions, LLM-based systems can be more reliable and trustworthy in real-world deployment.
• 🌏 Building Information-Seeking Tools With the Lowest Hallucination Rates: I also apply my research to build real-world information-seeking applications with the lowest hallucination rates, e.g., LingoWhale, RhinoInsight, and Zhiliao News released by DeepLang AI and TsinghuaNLP Lab. To date, these applications have provided trustworthy text processing services to hundreds of thousands of Chinese users.

I’m always happy to discuss potential collaborations. Feel free to drop an email if you are interested in connecting. 🕊🕊🕊

ssz24(at)mails.tsinghua.edu.cn Google Scholar GitHub