Chaeyun Jang

I'm a PhD student at KAIST AI, advised by Prof. Juho Lee.

Broadly, I'm interested in AI safety and alignment, with a particular focus on calibration.

What ultimately motivates this work is a form of inequality that comes not from access to AI, but from differences in how people are able to use it (Hargittai, 2002; Lee et al., 2026). As models become more capable, they can widen the gap between users who know how to test, question, and guide them toward deeper answers and users who receive more generic responses.

Calibration is one direction toward addressing this gap. I study how language models can better recognize and express what they do not know, so that their uncertainty helps users reason more carefully instead of simply making the model sound confident.

Anyone interested in discussing uncertainty calibration is always welcome to reach out.

jcy9911[at]kaist.ac.kr · Google Scholar · LinkedIn · X

Publications

No Pain, More Gain: Iterative Merging for Effective Multi-Teacher On-Policy Distillation

Seonghyeon Kim*, Chaeyun Jang*, Noah Lee, Boseop Kim, Juho Lee (*equal contribution)

Preprint

project page code

Mixture-Trained Merging for Unified Multi-Objective Models

SeongHyeon Kim*, Chaeyun Jang*, Seungyoo Lee, Jiyeon Ham, Yunju Bak, Boseop Kim, Juho Lee (*equal contribution)

NeurIPS 2026

to appear

Confidence is Not Universal: Task-Dependent Calibration and Emergent Behavior in LLMs

Chaeyun Jang, Moonseok Choi, Yegon Kim, Seungyoo Lee, Juho Lee†, Hyungi Lee† (†co-corresponding)

ICML 2026

Extends ICML 2025 R2-FM Workshop.

Bridging the Missing-Modality Gap: Improving Text-Only Calibration of Vision Language Models

Mingyeong Kim, Jungwon Choi, Chaeyun Jang, Juho Lee

ICLR 2026 Trustworthy AI Workshop

arxiv

Reliable Decision-Making via Calibration-Oriented Retrieval-Augmented Generation

Chaeyun Jang, Deukhwan Cho, Seanie Lee, Hyungi Lee, Juho Lee

NeurIPS 2025

arxiv code

Model Fusion through Bayesian Optimization in Language Model Fine-Tuning

Chaeyun Jang*, Hyungi Lee*, Jungtaek Kim†, Juho Lee† (*equal contribution, †co-corresponding)

NeurIPS 2024 Spotlight (top 2.1%, 327/15671)

arxiv code

Education

KAIST

Ph.D. in AI · advised by Prof. Juho Lee

2025 – Present

KAIST

M.S. in AI · advised by Prof. Juho Lee

2025

Sungkyunkwan University

B.S. in Statistics & Computer Science

2023

Experience

Part-Time Research Scientist

Kakao, Language Model Team · Seongnam, South Korea

Mar. 2026 – Present

Undergraduate Research Intern

Sungkyunkwan University · advised by JinYeong Bak

Jul. 2021 – Jul. 2022

Computer Vision Intern

Nuvilab Inc. · Food-tech AI startup, Seoul

Jun. 2020 – Dec. 2020

Teaching

Teaching Assistant

KAIST · AI708: Bayesian Machine Learning

Mar. 2024 – Dec. 2024

Talks

Reliable Decision-Making via Calibration-Oriented Retrieval-Augmented Generation

AI Tutorial Series, Seoul AI Hub · Seoul, South Korea

26 Sept. 2025

Tutorial Talk · poster

Model Fusion through Bayesian Optimization in Language Model Fine-Tuning

AI Technology Showcase 2025, KAIST Graduate School of AI · COEX, Seoul, South Korea

16 May 2025

Research Showcase Presentation · news

Academic Service

Conference Reviewer

NeurIPS, ICLR, ICML

Jul. 2024 – Present

Silver Reviewer Award ICML 2026