Physicist and Anthropic co-founder behind the neural scaling laws paper
A physicist, professor at Johns Hopkins and co-founder of Anthropic. His 2020 paper established the scaling laws: the observation that model performance improves predictably with more data, parameters and compute.
WikipediaWikipedia
#scaling laws#large language models
Key arguments & positions
Studies how model performance changes with training data, computing power and model size, and how those relationships can guide efficient frontier-model development.
Treats interpretability and model behavior as empirical research problems that should be investigated alongside capability gains.
Accomplishments
Co-founded Anthropic in 2021 and serves as its chief science officer.
Co-authored the 2020 neural scaling-laws paper that influenced how major labs plan model training.
Co-authored the GPT-3 paper and later contributed to Anthropic's work on Constitutional AI.
Associate professor of physics at Johns Hopkins University.
