Jihun Moon

I am a master's student in the Interdisciplinary Program in Cognitive Science at Seoul National University, advised by Prof. Youngjae Yu at the Physical & Interactive Intelligence Lab (SNU PI Lab). I received my B.S. in Mechanical Engineering from Seoul National University in August 2026.

My research lies at the intersection of computational linguistics and cognitive science. I view language models not as static input-output systems, but as dynamic language-processing systems that continuously update their internal states based on evidence gathered through interaction.

In particular, I am interested in the idea that language conveys not only explicitly encoded meaning, but also cues about implicit states such as a speaker's beliefs, intentions, and expectations. I study how linguistic and contextual evidence is integrated and competes during processing, how such evidence forms and updates internal states, and how these states in turn shape subsequent interpretation and language generation. In the longer term, I am also interested in how nonverbal signals such as speech, prosody, and gaze contribute to these interactional processes.

To investigate these questions, I draw inspiration from process-level hypotheses in psycholinguistics and neurolinguistics, and test alternative processing hypotheses by examining LLM behavior, internal representations, and causal interventions. Ultimately, my goal is to understand whether fundamentally different language-processing systems, such as humans and language models, converge on shared computational principles when solving the same linguistic problems.

Email  /  CV

Research

We introduce EGO, a framework in which a VLM selects a task-appropriate action within a predictive action boundary proposed by a V-JEPA2-based world model — aligning language reasoning with action possibilities grounded in visual experience, and improving both single-step embodied reasoning and multi-step procedure anticipation on egocentric video.

We introduce SPREAD, the first framework that enables a pre-trained world model to learn environmental changes online, and continuously.

Academic & Extracurricular Activities

Established Seoul National University's first student organization dedicated to humanoid robotics and physical AI as a founding president, leading the club's participation in the Humanoid Challenge held in June 2026.

Co-founded Roboticus, a non-profit student organization jointly launched by robotics clubs of KAIST and Seoul National University (MR, SHAPE, SIGMA), and served on the organizing staff of the 1st Robot Hackathon (Aug 2026), a student-planned 100-hour physical AI hackathon at KAIST where ten student teams built working robots from scratch.

Participated in the K-Robotics Startup Cup, a robotics startup competition hosted by KAIST, developing and pitching robotics-based business ideas.

Implemented an A*-based path planner with a feedback trajectory tracking controller, achieving the top performance among all teams.

Website template from Jon Barron.