Noah Lee

Noah.jpg

I am an LLM Researcher of the Kanana Team of Kakao, where I work on post-training LLMs. Previously, I graduated my Master’s degree at the Kim Jaechul Graduate School of AI of KAIST, jointly advised by James Thorne and Jinwoo Shin.

My current main research interest lies in (but not confined to):

  • Efficient post-training pipelines for LLM/LRMs
  • Bridging alignment and controllability for real-world use (meaningful reasoning, contextualized personalization, etc.)

Feel free to get in touch!


News

Jul 6, 2026 A paper has been accepted to ICML 2026.
Jan 25, 2026 MAPO has been accepted to AAAI 2026.
Sep 15, 2025 I joined the Kanana Team of Kakao Corp. to work on LLM post-training.
Jul 1, 2025 Two papers (Robust RM & AlphaPO) have been accepted to ICML 2025!
Jan 24, 2025 Our RM cross-lingual paper has been accepted to NAACL 2025!

Publications

  1. perpru.jpg
    Persona-Pruner: Sculpting Lightweight Models for Role-Playing
    Jinsu Kim, Jihoon Tack, Noah Lee, and 1 more author
    2026
    arXiv
  2. mapo.png
    Margin-aware Preference Optimization for Aligning Diffusion Models without Reference
    Jiwoo Hong*, Sayak Paul*, Noah Lee, and 3 more authors
    AAAI, 2026
    arXiv Code
  3. alphapo.png
    AlphaPO - Reward Shape Matters for LLM Alignment
    Aman Gupta, Shao Tang, Qingquan Song, and 8 more authors
    ICML, 2025
    arXiv
  4. robustrm.png
    On the Robustness of Reward Models for Language Model Alignment
    Jiwoo Hong, Noah Lee, Eunki Kim, and 5 more authors
    ICML, 2025
    arXiv
  5. cross.png
    Cross-lingual Transfer of Reward Models in Multilingual Alignment
    Jiwoo Hong*, Noah Lee*, Rodrigo Martínez-Castaño, and 2 more authors
    NAACL, 2025
    arXiv Code
  6. biggen.png
    The BiGGen Bench: A Principled Benchmark for Fine-grained Evaluation of Language Models with Language Models
    Seungone Kim, Juyoung Suk, Ji Yong Cho, and 29 more authors
    NAACL (Best Paper), 2025
    arXiv
  7. conseval.png
    Evaluating the Consistency of LLM Evaluators
    Noah Lee*, Jiwoo Hong*, and James Thorne
    COLING, 2025
    arXiv
  8. orpo.png
    ORPO: Monolithic Preference Optimization without Reference Model
    Jiwoo Hong, Noah Lee, and James Thorne
    EMNLP, 2024
    arXiv Code
  9. humllm.png
    Can Large Language Models Capture Dissenting Human Voices?
    Noah Lee*, Na Min An*, and James Thorne
    EMNLP, 2023
    arXiv Code