Ronen Eldan

Mathematics research

Ronen Eldan's work also spans pure mathematics, and the mathematics department of the Weizmann Institute of Science lists Eldan's prizes among its faculty honors. The Breakthrough Prize Foundation credits Eldan with creating the stochastic localization method, which led to progress on open problems in high-dimensional geometry and probability, including Bourgain's slicing problem and the KLS conjecture.¹¹¹²

Small language models

With Yuanzhi Li, Eldan wrote TinyStories, submitted in May 2023, which asks how small a language model can be while still producing coherent English. The paper's synthetic dataset of short stories and its family of small models are published on Eldan's Hugging Face account.¹²¹³³

Eldan co-authored Microsoft's phi-1 report in June 2023 and the phi-1.5 report in September 2023, and is listed among the contributors in Microsoft Research's December 2023 post on Phi-2. Eldan's name also appears on the Phi-3 report from April 2024 and the Phi-4 report from December 2024.^4514867

In October 2023, Eldan and Mark Russinovich posted "Who's Harry Potter?", which describes a way to make a language model forget a subset of its training data without retraining it from scratch.⁹

AI for science

Eldan is one of the authors of "Early science acceleration experiments with GPT-5," submitted in November 2025. The paper collects short case studies in mathematics, physics, astronomy, computer science, biology, and materials science in which the authors used GPT-5 in their ongoing research.¹⁰

Work and contributions

• Phi small language models Open weights

  Co-author of the phi-1 (2023), phi-1.5 (2023), Phi-3 (2024), and Phi-4 (2024) technical reports and a listed contributor on the Microsoft Research blog post announcing Phi-2. The phi-1.5 report states that the model was open-sourced, and its Hugging Face model card lists the MIT license.^4567814

• TinyStories Open data

  First author of the TinyStories paper (2023). The dataset card on Eldan's Hugging Face account lists the CDLA-Sharing-1.0 license, points to the paper, and names the models from the paper (TinyStories-1M through TinyStories-33M) published on the same account; the TinyStories-33M model card lists the MIT license.¹²¹³³

• Who's Harry Potter? Approximate Unlearning in LLMs Published research

  First author, with Mark Russinovich, of a 2023 paper proposing a technique for unlearning a subset of a language model's training data without retraining, demonstrated on Llama2-7b, with the fine-tuned model made available on Hugging Face.⁹

Timeline

1. May 12, 2023Submitted the TinyStories paper with Yuanzhi Li.¹
2. Jun 20, 2023Co-authored the phi-1 report, Textbooks Are All You Need.⁴
3. Sep 11, 2023Co-authored the phi-1.5 technical report.⁵
4. Oct 3, 2023Submitted "Who's Harry Potter?" on approximate unlearning in language models, with Mark Russinovich.⁹
5. Dec 12, 2023Listed as a contributor in Microsoft Research's announcement of Phi-2.⁸
6. Apr 22, 2024Co-authored the Phi-3 technical report.⁶
7. Dec 12, 2024Co-authored the Phi-4 technical report.⁷
8. Nov 20, 2025Co-authored "Early science acceleration experiments with GPT-5."¹⁰

Documented roles

The sources do not confirm a current role; these are the roles they document.

• Listed with an OpenAI affiliation on the November 2025 paper "Early science acceleration experiments with GPT-5", OpenAI¹⁰
• Listed with a Microsoft Research affiliation on the 2023 TinyStories and "Who's Harry Potter?" papers, Microsoft Research¹⁹
• Faculty member in the Faculty of Mathematics and Computer Science; promoted to Associate Professor with tenure in 2021, Weizmann Institute of Science¹²¹¹

Honors and recognition

• Erdős Prize in Mathematics · Israel Mathematical Union, 2018¹¹
• Blavatnik Award in Physical Sciences and Engineering · Blavatnik Family Foundation, 2022¹¹
• New Horizons in Mathematics Prize · Breakthrough Prize Foundation, 2023¹²

Companies in their history

• OpenAI 63 people in the wiki
• Microsoft 75 people in the wiki

Related people

• Ben MannConnected to OpenAI
• Susan ZhangConnected to OpenAI
• Jakub PachockiConnected to OpenAI
• Łukasz KaiserConnected to OpenAI
• Shunyu YaoConnected to OpenAI
• Hynek KydlíčekDatasets and data · Language models
• Pradeep DasigiLanguage models · Datasets and data
• Eugene YanLanguage models · Datasets and data
• Henrique Schechter VeraLanguage models · Datasets and data
• Vaishaal ShankarDatasets and data · Language models