PeopleofSuperintelligence

Researcher

Mark Sellke

Mark Sellke is a mathematician who works on high-dimensional probability, optimization and machine learning, and who studies what large language models can do for mathematics. He studied at MIT, Cambridge and Stanford, where his work on chasing convex bodies shared the SODA 2020 best paper award and, with Sébastien Bubeck, the Universal Law of Robustness won a NeurIPS 2021 Outstanding Paper Award. After 2022–2023 as a Member of the Institute for Advanced Study and a postdoctoral scientist at Amazon, he joined Harvard's statistics department in July 2023, and he received the 2025 Rollo Davidson Prize and the 2026 Bernoulli Society New Researcher Award. In 2025, he and Mehtaab Sawhney used GPT-5 to search the literature on conjectures of Paul Erdős, and in September 2026 they described results from OpenAI's reasoning models, including a sphere-packing bound. He is an assistant professor of statistics at Harvard and a researcher at OpenAI.

Badges

Timeline

  1. 2014–2022

    MIT, Cambridge and Stanford

    He studies mathematics at MIT in 2014–2017 and at the University of Cambridge in 2017–2018, then pursues a mathematics PhD at Stanford in 2018–2022 under Andrea Montanari. In 2020 and 2021, he visits the Simons Institute for the Theory of Computing as a graduate student.

  2. 2019–2020

    Chasing convex bodies optimally

    In May 2019, he posts “Chasing Convex Bodies Optimally” on arXiv, an online algorithm that reaches competitive ratio d in any normed space. The paper shares the SODA 2020 best paper award with a CMU team's work and earns him the best student paper award.

  3. 2021

    A universal law of robustness

    With Sébastien Bubeck, he proves in 2021 that smoothly interpolating data takes d times more parameters than merely interpolating it, where d is the ambient data dimension, a partial explanation for the heavy overparametrization of deep networks. Their paper “A Universal Law of Robustness via Isoperimetry” wins a NeurIPS 2021 Outstanding Paper Award.

  4. 2022–2023

    From the IAS to Harvard

    He is a Member of the Institute for Advanced Study in Princeton and a postdoctoral scientist at Amazon in 2022–2023. In July 2023, he joins Harvard as an assistant professor of statistics, where his research sits at the interface of probability, mathematical physics, theoretical computer science and statistics.

  5. 2025

    GPT-5 and Erdős problems

    In 2025, he and Mehtaab Sawhney use GPT-5 to search the literature on several conjectures of Paul Erdős in combinatorial number theory. In October 2025, his post listing Erdős problems for which GPT-5 found published solutions is misunderstood by OpenAI's Kevin Weil, who acknowledges the error and deletes his own post. In November 2025, he is one of fourteen authors of “Early science acceleration experiments with GPT-5,” a set of case studies across mathematics, physics, astronomy, computer science, biology and materials science.

  6. 2025–2026

    Rollo Davidson and Bernoulli prizes

    He shares the 2025 Rollo Davidson Prize with Hao Shen and Eliran Subag. He is one of three winners of the Bernoulli Society New Researcher Award 2026 in probability theory, with Morris Ang and Serte Donderwinkel.

  7. September 2026

    Sphere packing with OpenAI models

    In September 2026, he and Mehtaab Sawhney disclose that OpenAI's reasoning models helped resolve several long-standing open problems in pure mathematics, among them a high-dimensional sphere-packing bound that improves on the linear programming bound of the 1970s.

  8. As of October 2026

    Harvard and OpenAI

    He is an assistant professor of statistics at Harvard, working in high-dimensional probability, optimization and machine learning, and a researcher at OpenAI. In April 2026, he speaks at UMass Amherst on recent mathematical advances made by large language models.

Career

  1. Since 2023Harvard UniversityAssistant Professor
  2. NowOpenAIResearcher
  3. 2022–2023Institute for Advanced StudyMember
  4. 2022–2023AmazonPostdoctoral Scientist
  5. 2020–2021University of California, BerkeleyGraduate student visitor

Education

  1. 2022Stanford UniversityPhD
  2. 2018University of CambridgeMathematics
  3. 2017Massachusetts Institute of TechnologyMathematics

From the biography.

  1. 22 of 133 people shaping AI with a former OpenAI role now work at AnthropicIncludes Mark Sellke. Our estimates of where OpenAI recruits came from and where 133 people with former OpenAI roles work now, including the companies and labs they founded.
  2. The median h-index is 48 at universities and 14 at frontier AI labsIncludes Mark Sellke. The median h-index is 48 at universities and 14 at frontier AI labs. We compare publishing careers, recent citations, dual appointments and job moves across 2,420 university researchers and 308 frontier-lab researchers.
  3. 16% of AI career moves in the 2020s went from a university to a companyIncludes Mark Sellke. We estimate that 16% of AI career moves in the 2020s went from a university to a company; 29% of moves starting at a university went to a company, and 76% ended the university role.
  4. 23% of people shaping AI who left a university for a company later returned to a universityIncludes Mark Sellke. We estimate that 468 of the 2,011 people shaping AI who left a university for a company (23%) later returned to a university after a median of 5 years, and 163 of 829 (20%) made the reverse round trip.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Mark Sellke yet. Send us a free-licensed photo.

  1. Mark Sellke | alphaXiv alphaxiv.org. Checked 21 October 2026.
  2. Mark Sellke | Simons Institute for the Theory of Computing simons.berkeley.edu. Checked 21 October 2026.
  3. [1905.11968] Chasing Convex Bodies Optimally arxiv.org. Checked 21 October 2026.
  4. Convex body chasing, Steiner point, Sellke point, and SODA 2020 best papers | I’m a bandit blogs.princeton.edu. Checked 21 October 2026.
  5. NeurIPS 2021 Awards neurips.cc. Checked 21 October 2026.
  6. [2105.12806] A Universal Law of Robustness via Isoperimetry arxiv.org. Checked 21 October 2026.
  7. Mark Sellke msellke.com. Checked 21 October 2026.
  8. Mehtaab Sawhney - Wikipedia en.wikipedia.org. Checked 21 October 2026.
  9. OpenAI claimed a math breakthrough this weekend, only to be smacked down - Sherwood News sherwood.news. Checked 21 October 2026.
  10. [2511.16072] Early science acceleration experiments with GPT-5 arxiv.org. Checked 21 October 2026.
  11. Rollo Davidson Prize - Wikipedia en.wikipedia.org. Checked 21 October 2026.
  12. Bernoulli Society New Researcher Award bernoullisociety.org. Checked 21 October 2026.
  13. Research in the AI Era Seminar: Mark Sellke, Recent Mathematical Advances with Large Language Models : Manning College of Information & Computer Sciences : UMass Amherst cics.umass.edu. Checked 21 October 2026.