Jinyeop Song

Cambridge, MA, 02142

sjinyeop@gmail.com

617-949-1042

Hi, I’m Jinyeop Song, a 5th-year PhD student in the MIT Physics Department.

My research focuses on understanding the Science of AI, particularly on large language models (LLMs). Over the past few months, I’ve been exploring Neural Scaling Laws(ICL) and In-Context Learning(ICL), uncovering how these models adapt and grow in capability.

Another key interest of mine is building AI for Science. I’m currently involved in projects developing Scalable Foundational Models for multiomics and proteins, helping advance scientific discovery through machine learning.

I’m always open to discussions and collaborations—feel free to reach out to me!

news

Nov 06, 2025	Featured in MIT News for research on knowledge-graph augmented LLMs using multi-turn RL.
Sep 15, 2025	Two papers accepted to NeurIPS 2025 Workshops: Bayesian Optimization with LLMs and LLM Agent Empowerment.
Jun 01, 2025	Summer research intern at MIT-IBM Watson AI Lab.

latest posts

May 01, 2025	ICML Spotlight Paper Accepted
Jan 01, 2025	Started MATS 7.0 Cohort
Sep 01, 2024	MethylGPT Paper Archived and Under Review at Nature Method
Aug 01, 2024	Scaling Law Paper Accepted to TMLR

selected publications

Efficient and Transferable Agentic Knowledge Graph RAG via Reinforcement Learning

Jinyeop Song, Shuo Wang, Julian Shun, and 1 more author

2025

In Preparation

Bib

@article{song2025kgrag,
  title = {Efficient and Transferable Agentic Knowledge Graph RAG via Reinforcement Learning},
  author = {Song, Jinyeop and Wang, Shuo and Shun, Julian and Zhu, Yada},
  note = {In Preparation},
  year = {2025},
}

Estimating Empowerment of Language Model Agents

Jinyeop Song, Jeff Gore, and Max Kleiman-Weiner

arXiv preprint arXiv:2509.22504, 2025

Bib

@article{song2025empowerment,
  title = {Estimating Empowerment of Language Model Agents},
  author = {Song, Jinyeop and Gore, Jeff and Kleiman-Weiner, Max},
  journal = {arXiv preprint arXiv:2509.22504},
  year = {2025},
}

Context to Concept: Concept Encoding in In-context Learning

Jinyeop Song, Sumin Han, P. Argawal, and 1 more author

2024

ICML 2025 spotlight

Bib

@article{song2024context,
  title = {Context to Concept: Concept Encoding in In-context Learning},
  author = {Song, Jinyeop and Han, Sumin and Argawal, P. and Gore, Jeffrey},
  note = {ICML 2025 spotlight},
  year = {2024},
}