Intelligence Briefing
Founded the Center for AI Safety (CAIS) in 2022 with Oliver Zhang. Orchestrated the landmark May 2023 Statement on AI Risk, signed by hundreds of AI leaders including CEOs of OpenAI, Google DeepMind, and Anthropic, stating that mitigating extinction risk from AI should be a global priority. Created MMLU, the most widely used benchmark for evaluating LLM capabilities. Invented the GELU activation function, now the default in BERT, GPT, and Vision Transformers. Published "Introduction to AI Safety, Ethics, and Society" textbook (2024). Advisor to Scale AI.
BS, Computer Science — University of Chicago
PhD, Computer Science — University of California, Berkeley
Operational History
Advisor to Scale AI
Continues to serve as an advisor to Scale AI.
careerMay 2023 Statement on AI Risk
Orchestrated a statement signed by hundreds of AI leaders emphasizing the need to mitigate AI extinction risk.
policyMMLU Benchmark
Created the MMLU benchmark for evaluating large language model capabilities.
researchFounded Center for AI Safety
Co-founded the Center for AI Safety with Oliver Zhang.
foundingAGI Position Assessment
Unknown
Leading voice on AI existential risk. Believes advanced AI poses catastrophic and existential risks to humanity. Advocates for proactive safety research, robust evaluations, and governance frameworks.
Leading voice on AI existential risk. Believes advanced AI poses catastrophic and existential risks to humanity. Advocates for proactive safety research, robust evaluations, and governance frameworks.
Intercepted Communications
“Mitigating extinction risk from AI should be a global priority.”
“Advanced AI poses catastrophic and existential risks to humanity.”
“Robust evaluations and governance frameworks are essential for AI development.”
“The GELU activation function has transformed modern neural networks.”
“AI safety research must be proactive, not reactive.”
Research Output
Introduction to AI Safety, Ethics, and Society
2024Textbook covering foundational concepts in AI safety.
MMLU: A Benchmark for Evaluating Large Language Models
2023arXiv
Introduced a new benchmark for assessing LLM capabilities.
The GELU Activation Function
2023NeurIPS
Detailed the development and impact of the GELU activation function.
Known Associates
Organizational Affiliations
Current
Center for AI Safety
Founder
2022 - Present
Scale AI
Advisor
2023 - Present
Former
University of California, Berkeley
Researcher
2018 - 2022
Source Material
Dossier last updated: 2026-03-04