On May 19, Andrej Karpathy announced on X that he was joining Anthropic, stating, “The next few years will be a crucial formative period for cutting-edge research on large language models,” and expressing great excitement about “returning to research work.” According to TechCrunch, Karpathy officially started at Anthropic this week, joining the pre-training team led by Nick Joseph. Anthropic noted that he will also form a new team dedicated to accelerating pre-training research using Claude — a phase involving large-scale training that equips Claude with core knowledge and capabilities, and remains the most computationally intensive and resource-heavy stage in the entire LLM development process. TechCrunch described Karpathy as “one of the very few researchers capable of bridging LLM theory with large-scale training practices,” adding that Anthropic’s move signals its belief that AI-assisted research — rather than merely relying on raw computing power — is key to staying ahead of rivals like OpenAI and Google.
Karpathy co-founded OpenAI and from 2015 to 2017 focused on deep learning and computer vision research. He later joined Tesla to lead projects related to Autopilot and Full Self-Driving until 2022; after a brief stint back at OpenAI lasting a year, he left once more in 2024 to found Eureka Labs, an AI education startup. Through public courses such as “Neural Networks: Zero to Hero” and his YouTube channel, he enjoys widespread popularity among developers. He mentioned having “a deep passion for educational work and plans to keep advancing it when the time comes,” though the future direction of Eureka Labs remains unclear. On the same day, Anthropic also announced that cybersecurity veteran Chris Rohlf would join its frontier red team; previously active in Yahoo’s “Paranoid” security group and Meta, Rohlf will focus on leveraging AI to strengthen cyber defense mechanisms.