About
Hi, I'm Hyeonbin.
I’m a Master’s student at KAIST AI advised by Minjoon Seo
and currently doing internship at KRAFTON’s AI Companion Team.When I started grad school, I studied LLM reasoning as a route to AGI. Along the way I ran into the "bitter lesson": models often appear competent because they’ve seen similar training instances during training, and therefore performance will nevertheless saturate at the limits of the training distribution.
My work asks how to break this ceiling:
(1) creating (synthetic) data: leverage self-generated, feedback-rich experiences that reliably improve reasoning.
(2) invent new learning methods : design learning and evaluation paradigms that push models beyond superficial pattern matching, based on analysis of how these models actually learn.I’m always open to collaborations. Feel free to reach out!