ABAB Pedia

Dan Roberts

Overview

Contents10 sections
Key facts

Education: From optics at Duke to graduate study in Britain

At Duke he majored in physics and electrical and computer engineering and worked with David R. Smith on materials that change how light travels. A Goldwater Scholar and Duke University Faculty Scholar, he received a Marshall Scholarship in 2009, took Part III of the Mathematical Tripos at Cambridge and studied quantum information at Oxford.⁠[3][1]

Co-founding Diffeo

He co-founded Diffeo, a collaborative machine intelligence company, and became its chief technology officer. With colleagues in its research group he also organized a Text Retrieval Conference track on updating knowledge bases. His co-founders were John Frank and Max Kleiman-Weiner. All three were Hertz fellows and had met at the foundation’s summer workshop.⁠[1][3]

A physics doctorate and a cover designed by his sister

His MIT physics dissertation studied black holes, quantum chaos and computational complexity. Allan Wilfred Adams III supervised the thesis, which drew on four collaborations, including work with Douglas Stanford and Leonard SusskindLeonard SusskindLeonard Susskind is an American theoretical physicist and the Felix Bloch Professor of Theoretical Physics at Stanford University, regarded as one of the founders of string theory; he also advanced the holographic principle, black hole complementarity and the string theory landscape, and is known for The Theoretical Minimum courses and popular books.Open the full entry. He thanked his parents David and Linda, and his sister Nicole, who designed its cover at his request. He also thanked Kyle Mahowald, his roommate of six years, fellow roommate Robbie Kubala, and Max Kleiman-Weiner, who cultivated his interest in machine learning.⁠[2]

Princeton, New York and Salesforce

Publishing The Principles of Deep Learning Theory

On May 5 Cambridge University Press published The Principles of Deep Learning Theory, which he wrote with Sho Yaida. The book drew on theoretical physics to explain neural networks and grew out of research with Boris Hanin, whom the publisher credits as a contributor.⁠[5][1]

Leaving Sequoia and tracing his curiosity

An interview released on October 22 was recorded shortly before he left Sequoia; by publication he had joined OpenAI. He recalled constantly asking why as a child. His nineteen-month-old reminded him of that impulse, following a washing-machine repairman around and trying to peer inside the machine. While at Sequoia he also tried unsuccessfully to start another AI company.⁠[6][1]

Time to think before answering

In a Sequoia talk released on May 8 he described how performance could improve both with additional training compute and with more time spent on an answer. His subsequent biography identifies him as head of Frontiers after leading OpenAI’s Foundations of Reinforcement Learning team, and as a continuing visiting scientist at MIT’s Center for Theoretical Physics.⁠[7][1]

Related Organizations, Websites and Public Accounts

Diffeo: Co-founder and former CTO. Historical record: https://danintheory.com/ (opens in a new window).⁠[1]

Institute for Advanced Study: Postdoctoral researcher, School of Natural Sciences. Official website: https://www.ias.edu/ (opens in a new window).⁠[4]

Meta Platforms: Research scientist in the New York AI research team of the company then known as Facebook, Inc.. Official website: https://about.meta.com/ (opens in a new window).⁠[1]

Salesforce: Principal researcher. Official website: https://www.salesforce.com/ (opens in a new window).⁠[1]

OpenAI: Head of Frontiers. Official website: https://openai.com/ (opens in a new window).⁠[1]

Sources7 sources

  1. Dan Roberts • Theoretical Physics and AI (opens in a new window)danintheory.com↑
  2. CHAOS, Holography, And Other Science (opens in a new window)danintheory.com↑
  3. Daniel Roberts (opens in a new window)hertzfoundation.org↑
  4. Daniel Roberts (opens in a new window)ias.edu↑
  5. The Principles of Deep Learning Theory: An Effective Theory Approach to Understanding Neural Networks (opens in a new window)cambridge.org↑
  6. OpenAI Researcher Dan Roberts on What Physics Can Teach Us About AI (opens in a new window)youtube.com↑
  7. 9 Years to AGI? OpenAI’s Dan Roberts Reasons About Emulating Einstein (opens in a new window)youtube.com↑