Andrei Karpathy reveals details of his work at Anthropic

Andrei Karpathy has shared details about his move to Anthropic. He is working on pre-training research under Nick Joseph’s guidance. This is the most expensive and critical stage of AI development, defining the foundational capabilities of future Claude models.

Karpathy is building a team that will use Claude itself to accelerate research. The AI will become a full-fledged tool, designing experiments, analyzing results, and identifying patterns in training data.

Karpathy has put his educational projects, including the Eureka Labs startup, on temporary hold. Meanwhile, Anthropic has hired Chris Rolfe, a cybersecurity veteran from Meta, to strengthen the model testing team before releases.