Language models surface the unwritten code of science and society [Preprint]

Honglin Bao, Siyang Wu, Jiwoong Choi, Yingrong Mao, James A. Evans

The underlined names represent the master's students working with me.

TL; DR: We present an algorithmic framework -- pushing LLMs to speak out their heuristics by searching their deeper self-consistent hypotheses that can explain their decision-making. This framework surfaces the hidden codes embedded in many aspects of human society, including science, where an unwritten norm is that reviewers prioritize storytelling over intrinsic rigor.

Presented at ICSSI (Intl. Conf. on the Science of Science and Innovation) 2025. Slides

Introspective growth: Automatically advancing LLM expertise in technology judgment [Preprint]

Siyang Wu, Honglin Bao, Nadav Kunievsky, James A. Evans

The underlined names represent the master's students working with me.

TL; DR: Based on a new benchmark of 1.3 million hard-to-distinguish patent pairs, we study self-questioning as both a practical mechanism for automatically enhancing LLM comprehension of technologies and a diagnostic probe into how their internal and external knowledge are organized.

Hugging Face

Presented at Midwest Machine Learning Symposium 2025.

Where there's a will there's a way: ChatGPT is used more for science in countries where it is prohibited [Paper]

Quantitative Science Studies, 2025

Honglin Bao, Mengyi Sun, Misha Teplitskiy

TL; DR: After its release, ChatGPT was used for science more by countries where it was prohibited, with little evidence of its association with research quality.

X thread 1; X thread 2

Presented at Harvard D^3 Institute research workshop 2024.

From division to unity: A large-scale study on the emergence of Computational Social Science, 1990–2021 [Paper]

TheWebConf/WWW 2025

Honglin Bao, Jiawei Zhang, Mingxuan Cao, James A. Evans

TL; DR: We conducted the largest quantitative examination (on ~11 million papers) of how computational social science (CSS) emerges from— and transforms — the social sciences, and uncovered a division-to-unity pattern.

ACM Showcases; X thread; Bluesky thread

Interactive demo for CSS evolution using transformer embeddings

A simulation-based analysis of the impact of rhetorical citations in science [Paper]

Nature Communications, 2024

Honglin Bao, Misha Teplitskiy

TL; DR: Using a counterfactual simulation, we find that "bad" citing without intellectual influence reduces the reproduction of inequality in science.

Featured in Nature's Computational Social Science Collection; Selected media coverage “Swarm Agents Club 集智俱乐部”; X thread

Code and data

Presented at Computational Organization Modeling Society Brown Bag Seminar 2023 and Harvard D^3 Institute research workshop 2022. Slides

Persistence Paradox in Dynamic Science [Preprint]

Honglin Bao, Kai Li

TL; DR: Tracking how computer scientists responded to the unexpected success of AlexNet and the deep learning revolution it sparked, we highlight scientific breakthroughs as a mechanism for power reconfiguration within the field.

Presented at The 2024 Annual Workshop of the ASIS&T Special Interest Groups for Metrics and Scientific and Technical Information.

Cultural Ties in American Sociology [Preprint]

Revise & Resubmit, Poetics ("field top" in cultural sociology)

Alex Xiaoqin Yan, Honglin Bao, Tom R. Leppard, Andrew P. Davis

TL; DR: Status and geo-location shape "cultural ties" and the formation of schools of thought in American sociology.

X thread

Presented at ASA (Meeting of the American Sociological Association) Science Knowledge and Technology Section 2023, Harvard D^3 Institute research workshop 2023, and NC State Structures, Identities, and Society Seminar 2023. Slides