Minwoo (Josh) Kang

I’m a fifth-year PhD student in Computer Science at UC Berkeley, where I am affiliated with Berkeley AI Research (BAIR) and the SLICE Lab.
My research is advised by John Canny and John Wawrzynek.

My research spans various topics in language modeling and NLP, with applications in computational social science as well as language models as ‘‘agents’’.

Most recent interests:

• Language models as collaborative agents
  • How do we train models that infer beliefs & intentions of the user and generate language conditioned on latent intents for effective Human-AI collaboration?
• Language models as exploratory agents
  • How do we equip models with the ability to propose and test hypotheses, especially under environments with high uncertainty or underspecification of constraints, and effectively search over a space of viable solutions?
• Language models as models of human behavior
  • How do we condition models to approximate human users with high fidelity, and how should we rethink LLM post-training for such applications in studying human behavior?

For contact, please reach out to:
mkang at cs dot {my institution} dot edu