creators

Dan Hendrycks: Center for AI Safety director

Dan Hendrycks directs the Center for AI Safety, wrote the MMLU and MATH benchmarks, organised the 2023 Statement on AI Risk and advises xAI and Scale AI.

Dan Hendrycks is a machine learning researcher who co-founded and runs the Center for AI Safety (CAIS), a San Francisco nonprofit working to reduce societal-scale risks from AI; its own site gives his title as executive and research director. He holds a BS from the University of Chicago and a PhD in AI from UC Berkeley, where Dawn Song and Jacob Steinhardt supervised him. He is first author of MMLU, posted in September 2020, and of the MATH dataset, posted in March 2021, two benchmarks now used as standard tests of language models, and CAIS built Humanity's Last Exam with Scale AI, released on 23 January 2025 with 3,000 questions. He also contributed the GELU activation function used in BERT, GPT and vision transformers, and wrote the textbook 'Introduction to AI Safety, Ethics, and Society'. Through CAIS he organised the 2023 Statement on AI Risk, the one-sentence statement that mitigating extinction risk from AI should be a global priority alongside pandemics and nuclear war. He advises Elon Musk's xAI for $1 a year and Scale AI, announced in November 2024, for $12 a year, holding equity in neither. In March 2025 he published 'Superintelligence Strategy' with Eric Schmidt and Alexandr Wang.

Specialization

AI safety, model evaluation, machine learning robustness, AI policy

Country

United States