Junsu Kim

nurijunsu [at] gmail [dot] com

Hi, I am Junsu Kim.

Previously, I was a Student Researcher at Google DeepMind's Superhuman Reasoning team, working on post-training and inference-time methods for mathematical reasoning. We won a gold medal 🏅 at the International Mathematical Olympiad (IMO) with Gemini Deep Think, and built Aletheia, a research agent solving open problems in mathematics.

I'm interested in bringing these reasoning capabilities to real-world reasoning about people and our society, which requires much longer context and careful reward definition. Ultimately, I believe this development can be done in a way that benefits society, preserving human autonomy and self-actualization rather than replacing us.

My journey stems from the Ryu Optimization Group, where I first dived into the field of artificial intelligence and learned how to research. Before that, I enjoyed my time preparing for the IMO, where I won a silver medal and first learned how to think.

Currently I'm finishing my BS at Seoul National University, exploring ideas and people that can change the world together. If anything here resonates with you, schedule a 30-minute chat or reach out by email. I love meeting new people!

Projects

IMO gold-medal standard with Gemini Deep Think

Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad

Thang Luong, Edward Lockhart, Dawsen Hwang, Junehyuk Jung, Jonathan Lee, Nate Kushman, Pol Moreno, Yi Tay, Lei Yu, Golnaz Ghiasi, Garrett Bingham, Lalit Jain, Vincent Cohen-Addad, Theophane Weber, Ankesh Anand, Steven Zheng, Vinh Tran, Vinay Ramasesh, Andreas Kirsch, Jieming Mao, Zicheng Xu, Wilfried Bounsi, Vahab Mirrokni, Hoang Nguyen, Fred Zhang, Mahan Malihi, Yangsibo Huang, Yuri Chervonyi, Trieu Trinh, Junsu Kim, Mirek Olšák, Marcelo Menegali, Xiaomeng Yang, Richard Song, Miklós Z. Horváth, Aja Huang, and Goran Žužić.

Google DeepMind Blog

Accelerating Mathematical and Scientific Discovery with Gemini Deep Think

Tony Feng, Trieu H. Trinh, Garrett Bingham, Dawsen Hwang, Yuri Chervonyi, Junehyuk Jung, Joonkyung Lee, Carlo Pagano, Sang-hyun Kim, Federico Pasqualotto, Sergei Gukov, Jonathan N. Lee, Junsu Kim, Kaiying Hou, Golnaz Ghiasi, Yi Tay, YaGuang Li, Chenkai Kuang, Yuan Liu, Hanzhao (Maggie) Lin, Evan Zheran Liu, Nigamaa Nayakanti, Xiaomeng Yang, Heng-Tze Cheng, Demis Hassabis, Koray Kavukcuoglu, Quoc V. Le, Thang Luong

Google DeepMind Blog

Publications

LoRA Training Provably Converges to a Low-Rank Global Minimum or It Fails Loudly (But it Probably Won't Fail)

Junsu Kim, Jaeyeon Kim, Ernest K. Ryu

International Conference on Machine Learning (Oral Presentation), 2025

Clip-Low Increases Entropy and Clip-High Decreases Entropy in Reinforcement Learning of Large Language Models

Jaesung R. Park, Junsu Kim, Gyeongman Kim, Jinyoung Jo, Sean Choi, Jaewoong Cho, Ernest K. Ryu

arXiv Manuscript, 2025

Towards Robust Mathematical Reasoning

Thang Luong◊, Dawsen Hwang*, Hoang Nguyen*†, Golnaz Ghiasi*, Yuri Chervonyi*, Insuk Seo*, Junsu Kim*, Garrett Bingham, Jonathan Lee, Swaroop Mishra†, Alex Zhai, Clara Huiyi Hu, Henryk Michalewski, Jimin Kim†, Jeonghyun Ahn†, Junhwi Bae†, Xingyou Song, Trieu H. Trinh, Quoc V. Le, Junehyuk Jung◊

*Core and equal contributors

Empirical Methods in Natural Language Processing, 2025

Gold-medalist Performance in Solving Olympiad Geometry with AlphaGeometry2

Yuri Chervonyi, Trieu H. Trinh, Miroslav Olšák, Xiaomeng Yang, Hoang Nguyen, Marcelo Menegali, Junehyuk Jung, Junsu Kim, Vikas Verma, Quoc V. Le and Thang Luong

Journal of Machine Learning Research, 2025

Towards Autonomous Mathematics Research

Tony Feng, Trieu H. Trinh, Garrett Bingham, Dawsen Hwang, Yuri Chervonyi, Junehyuk Jung, Joonkyung Lee, Carlo Pagano, Sang-hyun Kim, Federico Pasqualotto, Sergei Gukov, Jonathan N. Lee, Junsu Kim, Kaiying Hou, Golnaz Ghiasi, Yi Tay, YaGuang Li, Chenkai Kuang, Yuan Liu, Hanzhao Lin, Evan Zheran Liu, Nigamaa Nayakanti, Xiaomeng Yang, Heng-Tze Cheng, Demis Hassabis, Koray Kavukcuoglu, Quoc V. Le, Thang Luong

arXiv preprint arXiv:2602.10177, 2026

Let's talk

I keep a daily slot open for a 30-minute chat with anyone, whether about research, the path here, or just to say hello.

Book a 30-minute chat →