Portrait
Jaeyong Ko
M.S. Student
Industrial Engineering
Seoul National University
Data Science & Business Analytics Lab
About Me

Hi! I am Jaeyong Ko, an M.S. student in Industrial Engineering at Seoul National University, advised by Prof. Pilsung Kang in the Data Science & Business Analytics Lab.

My research interests are in large language models, with a focus on mathematical reasoning, latent reasoning, and evaluation in the financial domain. I am currently interested in latent reasoning in vision-language models, especially how models reason beyond text-token reasoning traces.

Curriculum Vitae
Education
  • Seoul National University
    M.S. in Industrial Engineering
    Sep. 2025 - Present
  • Seoul National University
    B.S. in Industrial Engineering
    Mar. 2019 - Aug. 2025
Publications
Cliff Tokens: Identifying Single-Token Failure Triggers in LLM Mathematical Reasoning
Cliff Tokens: Identifying Single-Token Failure Triggers in LLM Mathematical Reasoning

Jaeyong Ko, Pilsung Kang, Yukyung Lee

arXiv preprint arXiv:2606.25524 2026

Identifies single-token triggers that can sharply change the success or failure of LLM mathematical reasoning, supporting fine-grained analysis of reasoning reliability.

Cliff Tokens: Identifying Single-Token Failure Triggers in LLM Mathematical Reasoning

Jaeyong Ko, Pilsung Kang, Yukyung Lee

arXiv preprint arXiv:2606.25524 2026

Identifies single-token triggers that can sharply change the success or failure of LLM mathematical reasoning, supporting fine-grained analysis of reasoning reliability.

All publications
Research Experience
  • Data Science & Business Analytics Lab, SNU
    Master's Student
    Sep. 2025 - Present
  • Ecole Polytechnique de Montreal
    Research Intern
    Apr. 2024 - Aug. 2024
    • Human-centric smart manufacturing workcells for Industry 5.0 applications.
  • Financial Engineering Lab, KAIST
    Research Intern
    Dec. 2023 - Feb. 2024
    • Reinforcement learning applications for goal-based wealth management.
Projects
  • Duo 2.0: Evaluating LLM Reliability in Long Financial Conversations
    Financial-domain conversational reliability benchmark and LLM-as-a-Judge evaluation.
    Granted by Kakaobank
    Dec. 2025 - Present
  • Development of Advanced Integrated Monitoring Technologies for Laser Welding
    Laser welding anomaly detection with time-series and image multi-modal data.
    Granted by LG Electronics
    Jul. 2025 - Dec. 2025
Other Experience
  • EXGRAM
    Product Manager Intern
    Jul. 2022 - Oct. 2022
    Planned app configurations and push notifications for Veloo.
  • SNU Startup Consulting Group (ACT)
    President & Team Member
    Mar. 2023 - Dec. 2023
    Led student consulting engagements and developed GTM strategies for startup clients.
  • Republic of Korea Army
    Corporal
    Jan. 2021 - Jul. 2022
Teaching
  • Peer Tutoring, SNU
    Course: Production Control (406.310)
    Sep. 2024 - Dec. 2024
  • Peer Tutoring, SNU
    Course: Human Factors Engineering (406.304)
    Sep. 2023 - Dec. 2023
News
2026
This homepage is now open. 🎉
Jun 27
New paper on arXiv: Cliff Tokens: Identifying Single-Token Failure Triggers in LLM Mathematical Reasoning
Jun 25