Researcher at Anthropic, previously at OpenAI, where he led work on scaling laws — the finding that model performance improves predictably with compute, data and size. Co-author of the GPT-3 paper.
Key arguments & positions
His research argues that language-model performance follows smooth, predictable "scaling laws" as compute, data and parameters grow — a finding that shaped the industry's bet on ever-larger models.
Accomplishments
Led the scaling-laws research program at OpenAI before moving to Anthropic.
Co-author of the GPT-3 paper and of the influential 2020 scaling-laws paper with Jared Kaplan.
