About
I'm Yuxuan Su, a master's student at the School of Social Sciences, Tsinghua University in Beijing, and I will start my Ph.D. here in 2027, working on AI for social science. I did my undergraduate degree here too, in History (B.A., 2025, GPA 3.97/4.0).
Alongside research, I run Entropy Order (ENTRO AI), a company that provides AI alignment data to LLM developers. We are backed by YC China, a well-known USD/RMB dual-currency fund, and reached about US$4 million in ARR, with frontier AI labs in China and abroad among our customers. The company is closely connected to my work on post-training data.
Research
AI for social science. I study concrete ways in which AI can accelerate discovery in the social sciences, and the conditions under which AI-assisted research is valid and reliable. My work so far covers LLM-based structuring and annotation of large corpora of political and historical texts, statistical inference that remains valid under run-to-run variability in model outputs, and multi-agent simulation for policy research.
Post-training data and environments. I study how post-training data and training environments affect model capability, in particular whether they extend what a model can do or only make existing capabilities more reliable. The goal is to characterize the boundary between the two, identify which kinds of data, tasks, and feedback signals move it, and develop ways to measure the difference.
Industry Experience
Aug. 2025 - present. Founder & CEO, Entropy Order. We build AI alignment data for LLM developers. Raised a US$300,000 seed round from YC China, a well-known USD/RMB dual-currency fund; about US$4 million ARR; customers include leading AI labs in China and abroad.
Publications
Peer-reviewed
- Tianguang Meng and Yuxuan Su. "A Large Language Model-Based Method for Classifying Political Historical Materials: From Historical Material Structuring to Intelligent Annotation." Fudan Journal (Social Sciences Edition) (复旦学报(社会科学版)), 68(2), 2026, 153-167. [In Chinese]
- Mingxuan Liu, Yuxuan Su, and Tianguang Meng. "Stratified Regulation: Unpacking Value Alignment of Chinese LLMs." Chinese Sociological Review, 2026, 1-33. DOI: 10.1080/21620555.2026.2714528
- Yuxuan Su. "Large Language Models Leading the Intelligent Computing Paradigm of Text Analysis." Chinese Social Sciences Today (中国社会科学报), forthcoming. [In Chinese]
Under review
- Yuxuan Su, Mingxuan Liu, and Tianguang Meng. "Decoding Value Alignment in Large Language Models: A Comparative Study of Cloud-Based and Local Deployments." Revise and resubmit, Journalism & Communication (新闻与传播研究). [In Chinese]
- Tianguang Meng and Yuxuan Su. "Policy Simulator: A New Paradigm for Policy Research under a Multi-Agent Framework." Revise and resubmit, Journal of Public Management (公共管理学报). [In Chinese]
Working papers
- Yuxuan Su and Tianguang Meng. "Large Language Models Empowering Text Analysis: Applications, Opportunities, and Challenges."
- Yuxuan Su and Jianjun Yu. "Robust Inference for AI-Powered Text Analysis under Output Fluctuations."
- Yuxuan Su and Xiaohong Yu. "Elite Networks and State Building: Power Sharing Dynamics in China, 1921-1976."
Education
- 2027 (incoming). Ph.D., Tsinghua University. Research direction: AI for social science.
- Aug. 2025 - June 2027. Master's student, School of Social Sciences, Tsinghua University.
- June 2025. B.A., History, Tsinghua University. GPA 3.97/4.0, top 5%.
Talks
- Social Science History Association Annual Conference, October 2025.
- American Political Science Association Annual Meeting, September 2025.
- 13th Beijing Youth Political Science Forum, June 2024 (one of two undergraduates selected).
- Tsinghua Youth League "Spark Style in My Eyes" interdisciplinary session, June 2024.
- 7th Workshop on Historical Politics, April 2024 (the only undergraduate selected).
- 8th Inter-generational Dialogue Academic Symposium, December 2023.
Awards
- 2026. National Scholarship, Ministry of Education of the People's Republic of China.
- 2025. Computational Social Science Incentive Program, Tsinghua University.
- 2024. Academic Excellence Scholarship, Tsinghua University.
- 2023. First-Class Scholarship for Comprehensive Excellence, Tsinghua University.
- 2023-2025. Spark Scientific and Technological Innovation Grant for Undergraduates, Tsinghua University, ¥10,000 (about US$1,400) (top 1%).
- 2022-2024. Research Grant for Undergraduate Student Projects, Tsinghua University, ¥30,000 (about US$4,200).
Teaching & Service
Teaching
- Summer 2026. Global Fieldwork Survey (全球田野调查, GDD), Tsinghua University (undergraduate).
University service
- June 2024 - present. Vice President, The 43rd Student Association of Science and Technology, Tsinghua University.
- Feb. 2024 - June 2024. Group leader, Science and Technology Innovation Center, Tsinghua Youth League Committee.
College service
- July 2024 - present. President, Student Association of Science and Technology, Zhishan College, Tsinghua University.
- June 2025 - present. Science and Technology Innovation Counselor, Zhishan College, Tsinghua University.
Association service
- June 2023 - present. Chairman, The 17th "Spark Project" Academic Talent Training Program, Tsinghua University.
- May 2023 - May 2024. President, Student Governance Modernization Research Association, Tsinghua University.
Membership
- Midwest Political Science Association, 2024-present.
- American Political Science Association, 2025-present.
Fun Facts
- I ran a half marathon in under 1 hour 30 minutes (April 16, 2025).
- I can do 27 consecutive overhand-grip pull-ups in a single set (May 31, 2026).
- I studied History as an undergraduate, and I still follow political science and the history of the Chinese Communist Party.
Contact
School of Social Sciences, Tsinghua University, Haidian District, Beijing 100084, China.
Code: github.com/Yuxuan-THU
关于我
我是苏宇轩,在清华大学社会科学学院读硕士,2027 年入学读博,方向是 AI for social science。本科也在清华,读的是历史学(2025 年毕业,GPA 3.97/4.0)。
做研究的同时,我在做熵基秩序(ENTRO AI),给大模型厂商提供 AI 对齐数据。公司获得了知名双币基金 YC China 的投资,ARR 约 400 万美元,国内外头部的前沿 AI 实验室都有合作。这家公司和我在后训练数据方面的研究关系密切。
研究
AI for social science。研究 AI 加速社会科学发现的具体路径,以及 AI 辅助研究在什么条件下有效、可靠。目前的工作包括:用大模型对大规模政治与历史文本做结构化和标注;在模型输出存在随机波动时保证统计推断的有效性;面向政策研究的多智能体模拟。
后训练数据与环境。研究后训练数据和训练环境如何影响模型能力,重点是它们究竟拓展了模型能做到的事,还是只让已有能力更稳定。目标是刻画二者的边界,找出哪些类型的数据、任务和反馈信号能推动这一边界,并发展相应的度量方法。
产业经历
2025.08 - 至今。创始人兼 CEO,北京熵基秩序科技有限公司。为大模型厂商提供 AI 对齐数据。获 YC China 30 万美元种子轮融资(知名双币基金);ARR 约 400 万美元;客户覆盖国内外头部 AI 实验室。
论文
已发表
- 孟天广、苏宇轩。“A Large Language Model-Based Method for Classifying Political Historical Materials: From Historical Material Structuring to Intelligent Annotation”。《复旦学报(社会科学版)》,2026 年第 68 卷第 2 期,153-167 页。[中文]
- Mingxuan Liu, Yuxuan Su, and Tianguang Meng. "Stratified Regulation: Unpacking Value Alignment of Chinese LLMs." Chinese Sociological Review, 2026, 1-33. DOI: 10.1080/21620555.2026.2714528
- 苏宇轩。“Large Language Models Leading the Intelligent Computing Paradigm of Text Analysis”。《中国社会科学报》,即将刊发。[中文]
在审
- 苏宇轩、Mingxuan Liu、孟天广。“Decoding Value Alignment in Large Language Models: A Comparative Study of Cloud-Based and Local Deployments”。《新闻与传播研究》,修改后再投。[中文]
- 孟天广、苏宇轩。“Policy Simulator: A New Paradigm for Policy Research under a Multi-Agent Framework”。《公共管理学报》,修改后再投。[中文]
工作论文
- Yuxuan Su and Tianguang Meng. "Large Language Models Empowering Text Analysis: Applications, Opportunities, and Challenges."
- Yuxuan Su and Jianjun Yu. "Robust Inference for AI-Powered Text Analysis under Output Fluctuations."
- Yuxuan Su and Xiaohong Yu. "Elite Networks and State Building: Power Sharing Dynamics in China, 1921-1976."
注:论文题目保留英文原题。
教育
- 2027(入学)。清华大学,博士。研究方向:AI for social science。
- 2025.08 - 2027.06。清华大学社会科学学院,硕士研究生。
- 2025.06。清华大学,历史学学士。GPA 3.97/4.0,专业前 5%。
报告
- 社会科学史学会(SSHA)年会,2025 年 10 月。
- 美国政治学会(APSA)年会,2025 年 9 月。
- 第十三届北京青年政治学论坛,2024 年 6 月(两位入选的本科生之一)。
- Tsinghua Youth League “Spark Style in My Eyes” 交叉学科专场,2024 年 6 月。
- 第七届历史政治学工作坊,2024 年 4 月(唯一入选的本科生)。
- 第八届代际对话学术研讨会,2023 年 12 月。
奖项
- 2026。国家奖学金,中华人民共和国教育部。
- 2025。计算社会科学奖励计划,清华大学。
- 2024。学业优秀奖学金,清华大学。
- 2023。综合优秀一等奖学金,清华大学。
- 2023-2025。“星火”本科生科技创新资助,清华大学,1 万元(约 1,400 美元)(前 1%)。
- 2022-2024。本科生科研项目资助,清华大学,3 万元(约 4,200 美元)。
教学与服务
教学
- 2026 年夏季。清华大学本科课程《全球田野调查》(GDD)。
校级服务
- 2024.06 - 至今。清华大学第四十三届学生科协副主席。
- 2024.02 - 2024.06。清华大学团委科技创新中心组长。
书院服务
- 2024.07 - 至今。清华大学致理书院学生科协主席。
- 2025.06 - 至今。清华大学致理书院科技创新辅导员。
学生组织服务
- 2023.06 - 至今。清华大学第十七期“星火计划”学术人才培养项目主席。
- 2023.05 - 2024.05。Student Governance Modernization Research Association 会长,清华大学。
学会会员
- 美国中西部政治学会,2024 年至今。
- 美国政治学会,2025 年至今。
一些有趣的事实
- 半程马拉松跑进了 1 小时 30 分钟(2025 年 4 月 16 日)。
- 正握引体向上,一组连续做过 27 个(2026 年 5 月 31 日)。
- 本科读历史学,至今仍关注政治学和中国共产党历史。
联系
北京市海淀区清华大学社会科学学院,100084。
代码:github.com/Yuxuan-THU