Sewon Min

Timeline

1. 2018

   Computer science at Seoul

   She earns a BS in computer science and engineering from Seoul National University in 2018.

2. 2020

   Answering ambiguous questions

   In 2020, she coauthors AmbigQA with Julian Michael, Hannaneh Hajishirzi and Luke Zettlemoyer. The task asks systems to find every plausible answer to a question and rewrite the question to make each interpretation clear.

3. 2022

   Testing in-context learning

   She coauthors Rethinking the Role of Demonstrations: What Makes In-Context Learning Work? in 2022. The experiments show that the set of possible labels, the distribution of input text and the format of examples help explain how language models perform tasks from examples in a prompt.

4. 2023

   Checking facts with FActScore

   In 2023, Min coauthors FActScore, a method for evaluating the factual accuracy of long text generated by language models. It divides an answer into individual facts and measures the share supported by a reliable knowledge source.

5. 2024

   A doctorate at Washington

   She completes her PhD in computer science and engineering at the University of Washington in 2024, advised by Hannaneh Hajishirzi and Luke Zettlemoyer in the UW NLP group. Her dissertation, Rethinking Data Use in Large Language Models, studies how models use text and develops models that retrieve information from a separate data store; during her doctorate, she also works as a part-time visiting researcher at Meta AI and interns at Google Research and Salesforce Research.

6. 2025

   The ACL dissertation award

   In 2025, she wins the ACL Computational Linguistics Doctoral Dissertation Award for Rethinking Data Use in Large Language Models. The award recognizes her work on language model behavior, including in-context learning.

7. July 2025

   Assistant professor at Berkeley

   Min joins UC Berkeley as an assistant professor in the Department of Electrical Engineering and Computer Sciences in July 2025.

8. 2025

   Training models with FlexOlmo

   In 2025, she leads research on FlexOlmo and develops it with Weijia Shi, Akshita Bhagia and Kevin Farhat. Data owners train expert modules on their own private data, combine them into a shared language model and control when their data is used.

9. May 2026

   Modular experts with EMO

   With Ryan Wang and Akshita Bhagia, she coauthors EMO: Pretraining Mixture of Experts for Emergent Modularity in May 2026. The model learns groups of experts that can be used separately for particular tasks; keeping 25% of its experts produces an absolute performance drop of about one percentage point.

10. As of October 2026

    Research scientist at Ai2

    As of October 2026, she is a research scientist at the Allen Institute for AI and an assistant professor in UC Berkeley EECS, affiliated with Berkeley AI Research and the Berkeley NLP Group. Her research centers on how large language models learn from and use data.