Dan Hendrycks is an American machine learning researcher whose work spans both model capabilities and AI safety. He earned a B.S. from the University of Chicago in 2018 and a Ph.D. in computer science from UC Berkeley in 2022.
Hendrycks has an unusual record of building artifacts that became field standards. As an undergraduate he co-authored the 2016 paper introducing the GELU (Gaussian Error Linear Unit) activation function, now used in most transformer models. In 2020 he led the creation of MMLU (Massive Multitask Language Understanding), which became the default broad-knowledge benchmark for large language models, and the MATH dataset of competition mathematics problems. He later co-created Humanity’s Last Exam, a deliberately extreme benchmark meant to outlast rapid model progress.
He is the founding director of the Center for AI Safety (CAIS), the San Francisco nonprofit that organized the 2023 Statement on AI Risk signed by leading researchers and executives. His research emphasizes catastrophic risk, robustness, and machine ethics. He also serves as a safety adviser to Elon Musk’s xAI and to Scale AI, in both cases reportedly for a symbolic one-dollar salary.