The recursive loop acquires a job title

Karpathy announced on 19 May 2026 that he had joined Anthropic, saying the next few years at the frontier of LLMs would be especially formative and that he wanted to get back to R&D (post); the newsletter adds that he reports to Nick Joseph and leads pre-training, which it glosses as training Claude to accelerate Claude. The position is recorded as a Role in this corpus, and it closes an arc that began with Claude running every optimization experiment on his nanochat project in December and his 650-experiment autoresearch loop in March. OpenAI’s Noam Brown answered that he would have loved for Karpathy to rejoin OpenAI but that the labs were ‘collectively advancing the most important tech of our era’ (post). The recursive-self-improvement storyline has moved from Altman’s production claim through an alignment lead’s present phenomenon and Clark’s 60% by 2028 to a named owner; by August a Google co-founder would be reported steering resources toward it.