# Dan Hendrycks

> Dan Hendrycks is the Executive Director of the Center for AI Safety (CAIS), the UC Berkeley PhD who created MMLU - the benchmark that shaped how the world measures AI intelligence - and invented GELU, the activation function running inside BERT, GPT, and virtually every major AI model. From a rural evangelical town in Missouri, he's become one of AI safety's most consequential voices, advising Elon Musk's xAI and Scale AI while sounding alarms about civilizational-scale AI risk, accepting $1/year to keep his independence intact.

- **Role:** Executive Director at Center for AI Safety (CAIS)
- **Organizations:** Center for AI Safety (CAIS), xAI (Adviser), Scale AI (Adviser), UC Berkeley (Alumnus)
- **From:** Marshfield, Missouri, USA
- **Nationality:** American
- **Education:** B.S. in Mathematics and Computer Science, University of Chicago, Ph.D. in Computer Science, UC Berkeley
- **Known for:** Created GELU activation function (2016) - now used in BERT, GPT, Vision Transformers, and virtually every major AI model, Created MMLU benchmark (2020) - 100M+ downloads, the standard for measuring AI reasoning across 57 subjects, Founded Center for AI Safety (CAIS) in 2022 - influential nonprofit shaping global AI safety discourse

## Career timeline

- **2016** — Published GELU (Gaussian Error Linear Units) paper as an undergraduate
- **2018** — Graduated University of Chicago; enrolled in UC Berkeley PhD program
- **2020** — Released MMLU benchmark - 15,908 questions across 57 subjects
- **2022** — Completed PhD at UC Berkeley; founded Center for AI Safety (CAIS)
- **2023** — Named AI Safety Adviser to xAI (Elon Musk's AI company); CAIS AI Risk Statement signed by 600+ experts; named to TIME 100 Most Influential People in AI
- **2024** — Co-founded Gray Swan AI; divested equity on principle; joined Scale AI as adviser
- **2025** — Co-authored Superintelligence Strategy with Eric Schmidt and Alexandr Wang; released Humanity's Last Exam (HLE); published MASK benchmark; spoke at Singapore Conference on AI

## Achievements

- Created GELU activation function (2016) - now used in BERT, GPT, Vision Transformers, and virtually every major AI model
- Created MMLU benchmark (2020) - 100M+ downloads, the standard for measuring AI reasoning across 57 subjects
- Founded Center for AI Safety (CAIS) in 2022 - influential nonprofit shaping global AI safety discourse
- Organized AI Risk Statement signed by 600+ AI scientists and leaders (2023)
- Named to TIME 100 Most Influential People in AI (2023)
- Schmidt Sciences AI2050 Early Career Fellow (2023)
- Co-authored Superintelligence Strategy with Eric Schmidt and Alexandr Wang (2025)
- Co-developed Humanity's Last Exam with Scale AI (2025)
- Published Introduction to AI Safety, Ethics, and Society textbook (2024)
- Influenced UK's first AI Safety Summit agenda
- NSF Graduate Research Fellowship recipient
- Open Philanthropy AI Fellowship recipient

## Latest updates

- **2025-07** — xAI signs safety portion of EU AI Act Code of Practice, announced by Hendrycks on X
- **2025-03** — Co-authored Superintelligence Strategy paper with Eric Schmidt and Alexandr Wang, introducing MAIM deterrence framework
- **2025-03** — Published MASK Benchmark showing frontier LLMs lie under pressure despite high truthfulness scores
- **2025-01** — Released Humanity's Last Exam (HLE) with Scale AI - 2,500 expert-level questions
- **2025-01** — Spoke at Singapore Conference on AI (SCAI) 2025
- **2024-11** — Joined Scale AI as adviser at $1/year salary
- **2024-10** — Released Remote Labor Index (RLI) - first benchmark of AI automation of real paid freelance work
- **2024-07** — Divested Gray Swan AI equity stake to avoid conflict of interest during SB 1047 debate

## Links

- Website: https://danhendrycks.com
- LinkedIn: https://www.linkedin.com/in/dan-hendrycks-04062395/
- Twitter/X: https://twitter.com/DanHendrycks
- GitHub: https://github.com/hendrycks

---

Profile page: https://yespress.io/dan-hendrycks
Published by YesPress — https://yespress.io
Last updated: 2026-04-27