Andrej Karpathy joins Anthropic pre-training

Andrej Karpathy said on May 19 that he had joined Anthropic, putting one of the best-known researchers from OpenAI’s early years and Tesla’s AI effort inside the startup’s pre-training organization. Karpathy disclosed the move in a post on X, writing that he was “very excited” to get back to research and development. Multiple outlets, including TechCrunch, CNBC and Bloomberg, reported that he is joining Anthropic’s pre-training team, which handles the large-scale training runs behind Claude. The move matters because pre-training is the stage where frontier labs spend heavily on data, compute and research talent to build a model’s base capabilities. Anthropic describes that group as responsible for the training runs that give Claude its core knowledge and capabilities, and reporting on Karpathy’s role says he started this week. TechCrunch and Yahoo’s aggregation of that report said the team is led by Nicholas Joseph, an early Anthropic hire and former OpenAI researcher. (techcrunch.com) ### What exactly did Karpathy say? Karpathy wrote on X on Tuesday: “I’ve joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative.” He added that he was “deeply passionate about education” and planned to resume that work over time. CNBC and TechCrunch both quoted the post in coverage published on May 19. May 19 was the date on which the move became public, and coverage across outlets described it as a return to hands-on research after Karpathy’s most recent stint outside a major lab. (msn.com) Bloomberg reported that he had left OpenAI for a second time last year and launched a startup focused on AI and education before taking the Anthropic role. ### What is Anthropic’s pre-training team responsible for? (cnbc.com) Anthropic’s pre-training organization works on the large-scale runs that create the base model before later tuning and deployment. TechCrunch’s report, echoed by Yahoo’s summary, said that is the group responsible for the runs that give Claude its core knowledge and capabilities. Another report said Karpathy’s mandate is to build a team focused on using Claude itself to accelerate pre-training research. (bloomberg.com) Pre-training is also one of the most compute-intensive parts of building a frontier model. MSN’s pickup of the TechCrunch report described the phase that way, underscoring why a hire in that unit drew attention beyond a routine personnel move. ### Why did this hire draw so much attention? Karpathy is a founding member of OpenAI and a former AI leader at Tesla, and those credentials were central to how the story spread on May 19 and May 20. (techcrunch.com) Forbes, CNBC and Bloomberg each highlighted his OpenAI and Tesla background in their first descriptions of the move. Axios described the hire as a major coup for Anthropic in the competition for elite AI talent, while Forbes framed it as a high-profile addition as labs race with OpenAI, Google and Meta for top researchers. (msn.com) Those are characterizations by those outlets, but they capture why the personnel change traveled quickly across tech and financial media. ### What does this say about the market for top AI researchers? (forbes.com) Anthropic’s decision to place Karpathy in pre-training shows where frontier labs are still concentrating scarce talent: core model development. Bloomberg reported that his work will focus on helping train new AI models, and TechCrunch said his remit sits inside the group responsible for Claude’s foundational training runs. (axios.com) Upstox, in a May 20 report, cast the move as part of the continuing AI talent war. That framing is from the publication, but it aligns with how several outlets positioned the hire—as a competition for a small pool of researchers with experience in both model architecture and large-scale training. ### What comes next inside Anthropic? (bloomberg.com) Nicholas Joseph is the named Anthropic leader attached to the pre-training team in current reporting, and Karpathy is reported to have started this week. The immediate next step is not a public product launch but research work inside the Claude training stack, including the effort to use Claude to speed pre-training research itself, according to reports describing his mandate. (forbes.com) Anthropic has not, in the reporting reviewed here, announced a dated product milestone tied to Karpathy’s arrival. The next concrete public signals are likely to come through future Anthropic model updates or any further statement from Karpathy about the team he is assembling inside pre-training. (techcrunch.com) (tech.yahoo.com)