Sam McCandlish

Sam McCandlish is an AI researcher and co-founder of Anthropic whose work concerns how large models are trained. He earned a doctorate in theoretical physics at Stanford University before moving into machine-learning research. His background connects quantitative investigation with the engineering and experimental challenges of training neural networks at scale.

At OpenAI, McCandlish worked on research into large-batch training and language-model scaling. He co-authored Scaling Laws for Neural Language Models in 2020, which examined how performance changed with model size, data, and computing resources. The work provided empirical guidance for thinking about training budgets and the balance among these resources.

McCandlish was also a contributor to the GPT-3 research published that year. In 2021, he joined colleagues in founding Anthropic, where research and infrastructure development are closely connected. Building increasingly capable systems requires decisions about training processes, computing arrangements, reliability, and the tools researchers use to conduct experiments.

Anthropic currently identifies him as its chief architect. The role includes large-scale training, pretraining research, researcher productivity, and reinforcement-learning infrastructure. His contribution is therefore both scientific and infrastructural: studying the behavior of training systems and helping create the machinery that makes experiments and model development possible. His current title reflects this broad technical scope rather than the earlier chief technology officer designation.