Andrej Karpathy joins Anthropic research

Andrej Karpathy said on May 19 that he has joined Anthropic, returning to hands-on research after prior stints at OpenAI and Tesla. (cnbc.com) He said the “next few years at the frontier of LLMs will be especially formative” and that he was excited to “get back to R&D.” (cnbc.com) Anthropic said Karpathy started this week on its pre-training team, the group responsible for the large-scale training runs that give Claude its core knowledge and capabilities. TechCrunch reported he is working under team lead Nick Joseph, and Joseph said Karpathy will build a team focused on using Claude to accelerate pre-training research itself. (cnbc.com) That makes the move notable for two separate reasons. First, Karpathy is one of the few researchers with senior experience across frontier lab research and large-scale deployment: he was a founding member of OpenAI, left in 2017 for Tesla, led Tesla’s computer vision work for Autopilot, returned to OpenAI in 2023, and left again in 2024. His personal site says that most recent OpenAI work involved midtraining and synthetic data generation, which overlaps with the kind of scaling work Anthropic is now assigning him. (cnbc.com) Second, Karpathy has an unusually large public footprint as an educator. His Stanford profile says he helped design and teach CS231n, described there as Stanford’s first deep learning course. His personal site says the class became one of Stanford’s most popular offerings. His “Neural Networks: Zero to Hero” course presents neural networks “from scratch, in code,” building up to GPT-style models, according to the course page. (cs.stanford.edu) Karpathy also said his education work is not over. In the same May 19 post, he wrote that he remains “deeply passionate about education” and plans to resume that work “in time.” TechCrunch noted he launched Eureka Labs in 2024 as an AI-native education startup, though it said it is unclear how that project fits with the Anthropic role. Eureka Labs’ site says the company was introduced in July 2024 to build an AI-native school. (techcrunch.com) The broader context is Anthropic’s continued buildup of research talent around Claude. CNBC reported Anthropic said Karpathy starts this week, while also noting other recent hires and infrastructure moves as competition with OpenAI and Google intensifies. Karpathy’s assignment is specific: not just working on a model, but helping use Claude itself to speed the research process behind future training runs. (cnbc.com) So the clearest read on this move is operational, not symbolic: Anthropic has put a high-profile researcher and teacher into one of the most compute-heavy, foundational parts of model development, and both Karpathy and Anthropic have said the work begins now. (cnbc.com)