Andrej Karpathy has joined Anthropic to return to R&D at the frontier of LLMs, while planning to resume education work later.
Key Takeaways
Karpathy left his independent period to rejoin frontier lab R&D, citing the next few years of LLM development as especially formative.
He retains a stated commitment to education work but frames it as deferred, not abandoned.
The move signals Anthropic continues to pull top-tier researchers away from independent or competing tracks.
Hacker News Comment Review
Karpathy is specifically joining Anthropic’s pretraining team, tasked with building a sub-team using Claude to accelerate pretraining research itself, extending ideas from his open-source autoresearch repo into a larger recursive training improvement effort.
Commenters noted Karpathy telegraphed this move in a recent interview, expressing concern about falling out of touch with frontier approaches and openness to rejoining a lab.
Some commenters raised concern about Anthropic’s accelerating talent consolidation, framing it as a structural risk to the broader ecosystem rather than just competitive positioning.
Notable Comments
@meetpateltech: Axios confirmed Karpathy starts this week on the pretraining team responsible for Claude’s core training runs.
@ollin: Anthropic pretraining lead confirmed he will build a team using Claude to accelerate pretraining research, linking to github.com/karpathy/autoresearch as the seed concept.
