Biography

KyoJun Choo is a student in Computer Science & Engineering at Soongsil University and a researcher in the Natural Language Processing Lab. His research focuses on Large Language Models (LLMs), particularly on understanding their internal representations and developing more reliable ways to evaluate their capabilities. His primary interests include LLM evaluation, benchmark reliability and validity, and representation-level interpretability, with additional interests in tool-augmented and agentic language models. His long-term goal is to contribute to AI systems that people can understand, evaluate, and use with greater confidence and trust. For more details, please see his CV.

Research Interests

• Large Language Model Evaluation
• Benchmark Reliability, Robustness, and Contamination
• Representation-level / Mechanistic Interpretability of LLMs

Education

• Feb. 2021 – Present: B.S. in Computer Science & Engineering, Soongsil University

Work Experiences

• Nov. 2025 – Present: Natural Language Processing Lab, Soongsil University, Undergraduate Researcher (Advisor: Chanjun Park)
• Mar. 2025 – Present: SH&M Coding Class, Spartan Future Education Institute, Coding Instructor

Publications

Conferences

• Structured but Silent: Probing Capability Requirements in LLM Hidden States
  Kyojun Choo, Minsoo Song, Yunju Kang, Chanjun Park^†
  [AACL-IJCNLP 2026 - Findings]

Domestic Conferences

• 행동적 타당성 감사를 통한 리더보드 강건성 분석 연구
  추교준, 송민수, 조성현, 장지윤, 남채린, 박찬준^†
  [HCLT 2026]
• 오답의 의미적 타당성 제거를 통한 LLM 벤치마크 결함 문항 식별
  송민수, 김찬우, 추교준, 송은서, 장지윤, 남채린, 이시현, 박찬준^†
  [HCLT 2026 - 최우수논문상]
• 많은 정보가 더 나은 예측을 만드는가? 대규모 언어 모델의 인간 어휘 이해도 예측 분석
  송은서, 조성현, 남채린, 추교준, 금지우, 박찬준^†
  [HCLT 2026]
• K-Press: 한국어 벤치마크 능력 프로파일링을 위한 문항 기반 능력 축 유도
  장지윤, 조성현, 김도윤, 송지훈, 추교준, 금지우, 박찬준^†
  [HCLT 2026]
• 한국어 상충 페르소나 지시에서의 LLM 해결 양상 분석: 동일 우선순위를 중심으로
  이시현, 금지우, 송지훈, 추교준, 김찬우, 박찬준^†
  [HCLT 2026]

Awards

• Best Paper Award (HCLT 2026)
  오답의 의미적 타당성 제거를 통한 LLM 벤치마크 결함 문항 식별
  송민수, 김찬우, 추교준, 송은서, 장지윤, 남채린, 이시현, 박찬준^†
  한글 및 한국어 정보처리 학술대회 (HCLT) 2026

Patents

Domestic Patent Applications

• Method and Apparatus for Determining Functional Requirements of User Queries Using Pre-Generation Hidden States of a Large Language Model
  대규모 언어 모델의 생성 전 은닉 상태를 이용한 사용자 질의의 기능적 요구사항 판단 방법 및 장치
  Chanjun Park, Kyojun Choo
  Korean Patent Application, filed on August 19, 2026 (Application No. 10-2026-0156645).
• Apparatus and Method for Auditing Performance Rankings among Artificial Intelligence Models Using Benchmark-Specific Audit Conditions
  벤치마크별 감사 조건을 이용한 인공지능 모델 간 성능 순위 감사 장치 및 방법
  Chanjun Park, Kyojun Choo
  Korean Patent Application, filed on October 6, 2026 (Application No. 10-2026-0190777).