PeopleofSuperintelligence

Lab leader

Dan Hendrycks

Dan Hendrycks is a machine learning researcher whose work includes GELU and MMLU, a 57-task test for text models. He earned a computer science PhD at UC Berkeley after a BS from the University of Chicago. His 2023 work surveys catastrophic AI risks and assesses GPT-4 and GPT-3.5 through DecodingTrust, a NeurIPS Outstanding Paper. He advised xAI in July 2023 while directing the Center for AI Safety. He developed a course whose 2024 textbook is Introduction to AI Safety, Ethics, and Society, and serves as Executive & Research Director of the Center for AI Safety.

Now at
Center for AI Safety
Works
103

Publication numbers from OpenAlex, as of 14 October 2026. Citations and h-index are hidden because this OpenAlex record may be incomplete or mix in other authors.

Photo of Dan Hendrycks

Timeline

  1. 2016

    Giving networks a new activation

    In 2016, Hendrycks and Kevin Gimpel propose Gaussian Error Linear Units (GELUs), an activation function that weights inputs by value. Their evaluation reports performance gains across computer vision, natural language processing, and speech tasks.

  2. 2018–2022

    From Chicago to Berkeley

    He earned a BS from the University of Chicago and completed a computer science PhD at UC Berkeley. His doctoral studies ran from 2018 to 2022.

  3. 2020

    Testing language understanding

    With six coauthors, he introduced Measuring Massive Multitask Language Understanding (MMLU), a test that covers 57 tasks and measures text models’ multitask accuracy.

  4. 2023

    Mapping risk and trust

    His paper with Mantas Mazeika and Thomas Woodside, An Overview of Catastrophic AI Risks, surveys sources of catastrophic AI risk. In DecodingTrust, he and his coauthors assess GPT-4 and GPT-3.5 across toxicity, bias, robustness, privacy, machine ethics, and fairness; NeurIPS names it an Outstanding Paper.

  5. July 2023

    Advising xAI

    At xAI’s July 2023 launch, he was an adviser while directing the Center for AI Safety.

  6. 2024

    Teaching AI safety

    He developed an AI safety course with an online textbook, Introduction to AI Safety, Ethics, and Society, published by Taylor & Francis in 2024. Its chapters address technical safety, machine ethics, collective action, and governance.

  7. As of October 2026

    Leading safety research

    He serves as Executive & Research Director at the Center for AI Safety, which works through research, field-building, and advocacy.

Sources and credits

Photo: TechCrunch, via Wikimedia Commons. License: CC BY 4.0. Source: commons.wikimedia.org. Cropped to a square and resized. Ask us to remove or replace this photo.

  1. [1606.08415] Gaussian Error Linear Units (GELUs) arxiv.org. Checked 14 October 2026.
  2. Dan Hendrycks – Center for Human-Compatible Artificial Intelligence chai.berkeley.edu. Checked 14 October 2026.
  3. Dan Hendrycks people.eecs.berkeley.edu. Checked 14 October 2026.
  4. [2009.03300] Measuring Massive Multitask Language Understanding arxiv.org. Checked 14 October 2026.
  5. [2306.12001] An Overview of Catastrophic AI Risks arxiv.org. Checked 14 October 2026.
  6. NeurIPS 2023 Awards neurips.cc. Checked 14 October 2026.
  7. [2306.11698] DecodingTrust: A Comprehensive Assessment of Trustworthiness in GPT Models arxiv.org. Checked 14 October 2026.
  8. Elon Musk wants to build AI to ‘understand the true nature of the universe’ techcrunch.com. Checked 14 October 2026.
  9. AI Safety, Ethics, and Society Textbook aisafetybook.com. Checked 14 October 2026.
  10. About Us | CAIS safe.ai. Checked 14 October 2026.