Caiqi Zhang | University of Cambridge

🙋 About Me

I am a third year Ph.D. student at the University of Cambridge, working under the supervision of Prof. Nigel Collier. Previously, I completed my MPhil degree at Cambridge, focusing on fact-checking under the guidance of Prof. Andreas Vlachos and Dr. Zhijiang Guo.

During my PhD studies, I did research internships in Microsoft Research, J.P. Morgan AI Research, and Tencent AI Lab.

During my undergraduate studies, I completed my capstone project with Prof. Wenjie Li on conversational QA systems and interned at UCLA under the supervision of Dr. Nanyun Peng.

🧐 Research Interests

Uncertainty in Large Language Models
- How can we quantify the uncertainty of LLMs for a given question?
- How can we teach the LLMs to proactively express uncertainties?
Factuality in Large Language Models: How can we reduce hallucinations in LLMs?

📚 Education

2023.10 - Present: University of Cambridge, Ph.D. in Computation, Cognition and Language.
2022.10 - 2023.06: University of Cambridge, M.Phil. in Advanced Computer Science.
2018.09 - 2022.06: The Hong Kong Polytechnic University, B.Sc. in Computing.
- Awarded the highest award for undergraduates: Outstanding Student of the Faculty of Engineering 2021/22. (Only 1 from more than 1000 students.) [News][Sharing][YouTube][Facebook][WeChat]

Undergraduate Scholarships:

• HKSAR Government Scholarship 2020/21 and 2021/22 (HKD 160,000, around USD 20,500)
• Commercial Radio 50th Anniversary Scholarship 2019/20 (HKD 80,000, around USD 10,250)
• The Hong Kong Polytechnic University Scholarship 2019/20 (HKD 40,000, around USD 5,125)
• Wong Tit-shing Student Exchange Scholarship 2020/21 (HKD 20,000, around USD 2,560)
• WKF Foundation Service-Learning Scholarship 2020/21 (HKD 16,600, around USD 2,125)
• Wei Lun Foundation Scholarship 2020/21 (HKD 16,600, around USD 2,125)
• Tellhow Group Scholarship 2018/19 (CNY 10,000, around USD 1,399)
• Rennie's Mill Student Aid Project Alumni Association Scholarship 2019/20 (HKD 10,000, around USD 1,250)
• V.K. Hsu & Sons Foundations Ltd. Scholarship 2019/20 (HKD 10,000, around USD 1,250)
• HKMA IT Management Club Scholarship 2021/22 (HKD 5,000, around USD 640)
• Proof-of-Concept (POC) Funding Scheme 2021/22 (HKD 5,000, around USD 640)

👨‍💻 Internships

Microsoft Research; Research Intern (Full-time); 2025
J.P. Morgan AI Research; Research Intern (Full-time); 3 months, 2025
- Work with Dr. Elizabeth Fons and Dr. Vamsi Potluru.
Tencent AI Lab; Research Intern (Full-time); 6 months, 2024
- Work with Dr. Zhizong Zhang and Dr. Xinting Huang.
PlusLab; University of California, Los Angeles; Research Intern (Part-time); 6 months, 2021
- Work with Dr. Nanyun Peng and Dr. Te-Lin Wu.
PolyU NLP Group; Research Assistant (Part-time); 12 months, 2021-2022
- Work with Prof. Maggie Wenjie Li and Dr. Yongqi Li.

🗺️ Research Roadmap

My research explores Uncertainty & Calibration in LLMs.

graph LR %% --- NODES & HIERARCHY --- %% Root Node Root(("🌲 Uncertainty
in LLMs")) %% --- Branch 1: Factuality --- Root --> Factuality("🔎 Factuality") %% Sub-category: Estimation Factuality --> L_Est["Post-hoc Estimation"] L_Est --> LUQ["📄 LUQ: First work on long-form UQ
(EMNLP '24)"] L_Est --> Atomic["⭐ Atomic Calibration
(IJCNLP '25)"] %% Sub-category: Expression Factuality --> L_Exp["Proactive Expression"] L_Exp --> LoGU["🗣️ LoGU: Linguistic Expressions
(ACL '25)"] L_Exp --> UNCLE["📏 UNCLE: Benchmarking
(EMNLP '25)"] L_Exp --> RL["🧠 RL for Verbalized Confidence
(Preprint)"] %% --- Branch 2: Reasoning --- Root --> Reasoning("🧩 Reasoning") Reasoning --> Rome["🗺️ All Roads Lead to Rome
(EMNLP '25)"] %% --- Branch 3: Multilingual --- Root --> Multilingual("🌐 Multilingual") Multilingual --> Beyond["🏗️ Beyond Final Layer
(Preprint)"] %% --- Branch 4: Multiturn --- Root --> Multiturn("💬 Multiturn") Multiturn --> Conformity["👥 Uncertainty leads to conformity
(ACL '25)"] Multiturn --> ConfMulti["🔄 Confidence Estimation fails in Multiturns
(Preprint)"] %% --- LINKS --- click LUQ "https://aclanthology.org/2024.emnlp-main.299/" "View Paper" click Atomic "https://arxiv.org/abs/2410.13246" "View Paper" click LoGU "https://arxiv.org/abs/2410.14309" "View Paper" click UNCLE "https://arxiv.org/abs/2505.16922" "View Paper" click RL "https://arxiv.org/abs/2505.23912" "View Paper" click Beyond "https://www.arxiv.org/abs/2510.03136" "View Paper" click Rome "https://arxiv.org/abs/2509.12908" "View Paper" click Conformity "https://arxiv.org/abs/2410.12428" "View Paper" %% --- STYLING --- classDef main fill:#ffffff,stroke:#03396c,stroke-width:2px,color:white,font-size:18px; classDef domain fill:#ffffff,stroke:#03396c,stroke-width:2px,rx:10,ry:10,color:#03396c; classDef label fill:#fff,stroke:none,color:#666,font-size:15px; classDef paper fill:#fff,stroke:#ddd,stroke-width:1px,rx:5,ry:5,color:#333; %% Apply Classes class Root main; class Factuality,Reasoning,Multilingual,Multiturn domain; class L_Est,L_Exp label; class LUQ,LoGU,UNCLE,RL,Rome,Beyond,ConfMulti,Atomic,Conformity paper;

📝 Publications

† denotes equal contribution.

Core Research: Uncertainty in LLMs

Beyond the Final Layer: Intermediate Representations for Better Multilingual Calibration in Large Language Models
Preprint
Ej Zhou, Caiqi Zhang†, Tiancheng Hu, Chengzu Li, Nigel Collier, Ivan Vulić, Anna Korhonen
Reinforcement Learning for Better Verbalized Confidence in Long-Form Generation
Preprint
Caiqi Zhang†, Xiaochen Zhu†, Chengzu Li, Nigel Collier, Andreas Vlachos
UNCLE: Benchmarking Uncertainty Expressions in Long-Form Generation
EMNLP 2025 Main
Ruihan Yang†, Caiqi Zhang†, Zhisong Zhang, Xinting Huang, Dong Yu, Nigel Collier, Deqing Yang
All Roads Lead to Rome: Graph-Based Confidence Estimation for LLM Reasoning
EMNLP 2025 Main
Caiqi Zhang, Chang Shu, Ehsan Shareghi, Nigel Collier
Conformity in Large Language Models
ACL 2025 Main
Xiaochen Zhu†, Caiqi Zhang†, Tom Stafford, Nigel Collier, Andreas Vlachos
LoGU: Long-form Generation with Uncertainty Expressions
ACL 2025 Main
Ruihan Yang†, Caiqi Zhang†, Zhisong Zhang, Xinting Huang, Sen Yang, Nigel Collier, Dong Yu, Deqing Yang
Atomic Calibration of LLMs in Long-Form Generations
ACL 2025 KnowFM Oral / AACL-IJNLP 2025
Caiqi Zhang, Ruihan Yang, Zhisong Zhang, Xinting Huang, Sen Yang, Dong Yu, Nigel Collier
LUQ: Long-text Uncertainty Quantification for LLMs
EMNLP 2024 Main
Caiqi Zhang, Fangyu Liu, Marco Basaldella, Nigel Collier.

Other First & Co-First Papers

Do We Need Language-Specific Fact-Checking Models? The Case of Chinese
EMNLP 2024 Main
Caiqi Zhang, Zhijiang Guo, Andreas Vlachos.
TopViewRS: Vision-Language Models as Top-View Spatial Reasoners
EMNLP 2024 Main (Oral)
Chengzu Li†, Caiqi Zhang†, Han Zhou, Nigel Collier, Anna Korhonen, Ivan Vulić.

Other Collaborations (Full List: Google Scholar)

Can Large Language Models Generate High-quality Patent Claims?
NAACL 2025 Findings
Lekang Jiang, Caiqi Zhang, Pascal A Scherz, Stephan Goetz
Language is All a Graph Needs
EACL 2024 Findings
Ruosong Ye, Caiqi Zhang, Runhui Wang, Shuyuan Xu, Yongfeng Zhang.
Learning to Infer Action-Condition Dependencies from Instructional Manuals for Structural Instruction Understanding
ACL 2023 Main
Te-Lin Wu, Caiqi Zhang, Carol Hu, Alex Spangher, Nanyun (Violet) Peng.

👀 More facts about me:

Volunteer Teaching

During term breaks, I volunteered in various teaching trips to rural areas globally, covering Hong Kong, Taiwan, Guilin, Ho Chi Minh City (Vietnam), Phnom Penh (Cambodia), and Trà Vinh (Cambodia). I've participated in 10+ voluntary services, accumulating 400+ service hours, benefiting 300+ students. Also, I joined the United Nations' Millennium Fellowship 2021 to promote equal education.

Mandarin Debate

As a member of both the PolyU and Cambridge Mandarin Debate Teams, I participated in competitions across various cities, including Singapore, Shanghai, Suzhou, Nanjing, Wuhan, Changsha, Xi'an, and Chengdu. These experiences refined my communication and critical thinking skills and provided international representation opportunities.

Less is more. -Ludwig Mies Van der Rohe