Donghyeon Shin

Donghyeon Shin

AI Researcher at LG AI Research

Former M.S. Graduate from GIST (Feb 2026)

Major: Artificial Intelligence | Minor: Mathematics

Mail | CV | GitHub | Google Scholar

Hello! I am an AI Researcher at LG AI Research, where I contribute to developing world-class Large Language Models (LLMs). My dream is the universalization of intelligence through the development of Artificial General Intelligence (AGI). To achieve this goal, my research interests are focused on value alignment, safety, and reasoning.

I recently graduated with a Master's degree in Artificial Intelligence from GIST (February 2026), where I conducted research at the Data Science Laboratory under the supervision of Professor Sundong Kim. My work centered on LLMs, exploring topics including reinforcement learning, the analysis of LLMs using Sparse Autoencoders, and developing data augmentation and learning frameworks.

Work Experience

LG AI Research - AI Researcher

March 2026 - Present

  • Contributing to the development of world-class Large Language Models (LLMs) as part of the World Best LLM Team
  • Conducting research on advanced LLM capabilities, reasoning, and alignment

Education

Gwangju Institute of Science and Technology (GIST)

February 2018 - February 2026

  • B.S Major: Electrical Engineering and Computer Science
  • M.S Major: Artificial Intelligence
  • Minor: Mathematics
University of California, Berkeley (UC Berkeley)

June 2019 - August 2019

  • Coursework: Discrete Mathematics, Introduction to Probability and Statistics

Research Experience

   
Data Science Lab - Graduate Researcher
   

March 2023 - February 2026

   

Supervised by Prof. SunDong Kim

   
         
  • Investigated LLM reasoning capabilities by applying reinforcement learning, test-time strategies, and multi-modal frameworks to benchmarks like the Abstraction and Reasoning Corpus (ARC).
    •      
  • Analyzed the abstract reasoning abilities of LLMs using the ARC benchmark (Published in ACM TIST 2025).
    •      
  • Developed and proposed the MC-LARC benchmark to measure LLM reasoning (Published in EMNLP Findings 2024).
    •      
  • Examined the emergence of gambling addiction symptoms in LLMs, drawing parallels with human psychology
    •    

Publications

Tracing and Correcting Programs: Critic-Guided Synthesis for Visual Reasoning

Marha Midhatiey Rusli, Donghyeon Shin, Sejin Kim, and Sundong Kim

AAAI-26 Bridge Program: Logical and Symbolic Reasoning in Language Models (2026)

Can Large Language Models Develop Gambling Addiction?

Seungpil Lee, Donghyeon Shin, Yunjeon Lee, and Sundong Kim

arXiv:2509.22818 | Paper

Reasoning Abilities of Large Language Models: In-Depth Analysis on the Abstraction and Reasoning Corpus

Seungpil Lee*, Woochang Sim*, Donghyeon Shin*, Sanha Hwang, Wongyu Seo, Jiwon Park, Seokki Lee, Sejin Kim, and Sundong Kim

ACM TIST (2025) | Paper | Code | Website

From Generation to Selection: Findings of converting Analogical Problem-Solving into Multiple-Choice Questions

Donghyeon Shin*, Seungpil Lee*, Klea Lena Kovačec, and Sundong Kim

EMNLP Findings (2024) | Paper | Website

Regulation Using Large Language Models to Generate Synthetic Data for Evaluating Analogical Ability

Donghyeon Shin, Seungpil Lee, Klea Lena Kovačec, and Sundong Kim

IJCAI Workshop (2024) | Paper

Walk-on-Hemispheres First-Passage Algorithm

Jinseong Son, Donghyeon Shin, and Chi-Ok Hwang

Scientific Reports (2023) | Paper