Seungju Han
I am a predoctoral researcher working with Yejin Choi. Previously, I was a visiting student researcher at Allen Institute for AI (Ai2), also advised by Yejin Choi. At Ai2, I collaborated with Nouha Dziri, Youngjae Yu, and Jack Hessel. I received my B.S. in Electrical and Computer Engineering from Seoul National University and worked on research and engineering at Hyperconnect (a startup acquired by Match Group for $1.7B).
My research focuses on natural language processing and machine learning, particularly interested in scalable and practical ways to improve models. This has led me to work on:
- training models with scale: I built diverse, high-quality, large-scale datasets. For example, I employed internet-scale (20M videos) to teach vision-grounded dialogue (CHAMPAGNE), and leveraged in-the-wild user-LLM interactions to build 262K prompt-response pairs for red teaming (WildTeaming).
- evaluating models with models: I developed model-based evaluations, such as detecting the harms and refusals in user-LLM interactions (WildGuard) and measuring the diversity of LM responses. I also created model-written benchmarks, such as personality test to assess LLM behavior and challenging vision-language benchmarks (understanding social norms, humor, and visual arguments).
- building small deployable models: I worked on an algorithm for knowledge distillation from generative LMs to retrieval-based LMs, and retrieval-based LMs for in-context learning and response generation for practical chatbot models.
You can reach me at wade3han at snu.ac.kr.