Hello world! 👋

I’m Minwu, a researcher at Seoul National University, advised by Youngjae Yu. I am working on building world-event forecasting models/systems.

Before that, I was in New York University Abu Dhabi, where I was fortunate to be advised by Keith Ross, working on LLM reasoning. I also completed my undergraduate degree here, majoring in Computer Science.

Before that, I worked in computational finance as well, particularly in corporate finance and macroeconomics.

Before that, I worked on a startup project.

For a detailed look at my background and experiences, feel free to check out my CV. If you would like to talk about research or potential collaboration, feel free to ping me at minwukim[at]snu[dot]ac[dot]kr :)

News

07.2026 Joining SNU. I will continue working on LLM reasoning & world-event forecasting.
01.2026 New paper Training Reasoning Models on Saturated Problems via Failure-Prefix Conditioning is out.
08.2025: One work is accepted at EMNLP 2025 Main Conference.
06.2025: Preprint for my new paper Layer Importance for Mathematical Reasoning is Forged in Pre-Training and Invariant after Post-Training is out.
05.2025: Preprint for my new paper Reinforcement Learning vs. Distillation: Understanding Accuracy and Capability in LLM Reasoning is out.
05.2025: Preprint for my new paper Warm Up Before You Train: Unlocking General Reasoning in Resource-Constrained Settings is out.
02.2025: Preprint for my new paper Mathematical Reasoning in Large Language Models: Assessing Logical and Arithmetic Errors across Wide Numerical Ranges is out!
10.2024: My paper Interpretable Machine Learning Model for Predicting Activist Investment Targets is published at The Journal of Finance and Data Science.
09.2024: Started working as a research assistant at NYUAD.

Publications & Preprints

[7] Training Reasoning Models on Saturated Problems via Failure-Prefix Conditioning (link)
Minwu Kim, Safal Shrestha, and Keith Ross
Preprint, under review, 2026.

[6] On the Limits of Layer Pruning for Generative Reasoning in LLMs (link)
Safal Shrestha*, Anubhav Shrestha*, Aadim Nepal, Minwu Kim, and Keith Ross
Preprint, under review, 2026.

[5] Reinforcement Learning vs. Distillation: Understanding Accuracy and Capability in LLM Reasoning (link)
Minwu Kim*, Anubhav Shrestha*, Safal Shrestha, Aadim Nepal, and Keith Ross
MATH-AI @ NeurIPS 2025.

[4] Warm Up Before You Train: Unlocking General Reasoning in Resource-Constrained Settings (link)
Safal Shrestha, Minwu Kim, Aadim Nepal, Anubhav Shrestha, and Keith Ross
EMNLP 2025 Main Conference.

[3] Layer Importance for Mathematical Reasoning is Forged in Pre-Training and Invariant after Post-Training (link)
Aadim Nepal, Safal Shrestha, Anubhav Shrestha, Minwu Kim, Jalal Naghiyev, Ravid Shwartz-Ziv, and Keith Ross
MATH-AI @ NeurIPS 2025; BlackboxNLP @ EMNLP 2025.

[2] Mathematical Reasoning in Large Language Models: Assessing Logical and Arithmetic Errors across Wide Numerical Ranges (link)
Safal Shrestha*, Minwu Kim*, and Keith Ross
Preprint.

[1] Interpretable Machine Learning Model for Predicting Activist Investment Targets (link)
Minwu Kim, Sidahmed Benabderrahmane, and Talal Rahwan
The Journal of Finance and Data Science, 2024