Junseok Kim (Andrew)

Hello 👋! I am a Ph.D student at Seoul National University, advised by Prof. Kyomin Jung. My research lies at the intersection of Interpretability and AI Safety, asking how large language models work internally, and using those insights to make them safer and more aligned.

Email  /  Google Scholar  /  Github  /  Linkedin

News

Research

My research asks two interconnected questions: 1) How do large language models actually work? and 2) How can we make them more reliable and aligned? I use mechanistic interpretability to trace how internal representations and attention mechanisms shape model behavior — including when and why models express unwarranted confidence. I then leverage these insights to design inference-time strategies that regulate model decisions based on reliability signals, with the broader goal of building AI systems that are trustworthy by design.

We propose Personalized Privacy Control in LLMs via Attention Head Intervention, a method that allows users to control the privacy of their interactions with large language models by intervening in the attention mechanisms.

We propose Tool Selection Bias Amplifies in Multi-turn User–Agent Interactions, a study that examines how tool selection bias can amplify in multi-turn interactions between users and agents.

We propose Reliability-Aware Adaptive Self-Consistency (ReASC), an adaptive self-consistency framework that incorporates a response-level confidence as a reliability signal to guide how evidence is accumulated at inference time.

Persona Switch is a training-free decoding method that improves reasoning by step-wise switching between zero-shot and role-play prompting based on token-level confidence at decoding time.

We analyze the impact of role-play prompts in zero-shot reasoning tasks and show that they can be detrimental to performance in some cases depending on the designed persona.

We introduces "Decomposition of Experts" (DoE), a framework that accelerates inference by dynamically identifying and activating only task-specific neurons within a language model to reduce computational costs without sacrificing accuracy.

Last updated in August 2026. This page is based on Jon Barron's website template.