Andrej Karpathy Joins Anthropic to Lead Claude Pre-Training Acceleration

Summary

Andrej Karpathy, a founding member of OpenAI and one of the most prominent AI researchers of the past decade, has joined Anthropic to lead a team focused on accelerating Claude's pre-training research. Announced on May 19, Karpathy started immediately on Anthropic's pre-training team—the group responsible for the foundational training runs that produce Claude's base models before fine-tuning. In a significant twist, his team will focus on using Claude itself to accelerate pre-training research, exemplifying an industry-wide trend where leading labs deploy their own models to design experiments, write training code, and analyze data.

The hire represents a major talent win for Anthropic in a highly competitive market where experienced pre-training researchers command premium attention. Karpathy's resume is extraordinary: OpenAI co-founder (2015), Tesla's Autopilot and Full Self-Driving lead (2017–2022), AGI researcher at OpenAI (2023), and founder of Eureka Labs, an AI-native education startup that is being paused rather than shuttered. His recruitment continues Anthropic's sustained talent drain from OpenAI, which has already lured co-founders Dario and Daniela Amodei, chief scientist Jared Kaplan, and numerous lead researchers behind GPT-2 and GPT-3.

The move signals a broader shift in where pre-training research momentum is concentrating and underscores the premium placed on experienced researchers who understand training at scale. While Karpathy's immediate focus is advancing Claude, Anthropic has preserved his ability to resume educational work through Eureka Labs later, suggesting his role may evolve as Anthropic's pre-training capabilities mature.

• Karpathy is pausing (not closing) Eureka Labs, his education startup, suggesting his Anthropic tenure may be time-bounded