Jaehoon Kim

• I’m an MS student in AI at Yonsei University, advised by Professor Dongha Lee in the DLI Lab. Previously, I received my BS in Physics and Computer Science from Sogang University.
• I’m interested in anything that helps thinking machines think deeper.
  • To my current understanding, this includes
    • instilling reasoning abilities through supervised fine-tuning on reasoning traces that exhibit proper exploration and reflection in their thought processes, as well as
    • eliciting reasoning capacities latent within the models through reinforcement learning from experience in carefully designed interactive environments.
  • My work lies at the intersection of these approaches, developing demonstrations that align with how a learner naturally reasons and learning environments that provide richer guidance throughout the reasoning process, with inspiration from the cognitive processes of the human mind.
• ✨ hydrogen and helium → ⭐ stars → 🌏 earth → 👫 humans → 👣 giants → ⚡🧠 intelligence
• Email / GitHub / LinkedIn / Google Scholar

Publications

• Hindsight Compacts but Does Not Repair: Rethinking On-Policy Self-Distillation in Reasoning Models
  • Jaehoon Kim, Dongha Lee
  • Preprint (arXiv), 2026
  • paper / code
• In Their Own Words: Reasoning Traces Tailored for Small Models Make Them Better Reasoners
  • Jaehoon Kim, Kwangwook Seo, Dongha Lee
  • Preprint (arXiv), 2025
  • paper / code