All AI Updates
IndustryMay 19, 2026Anthropic

Andrej Karpathy Joins Anthropic's Pre-Training Team

The OpenAI founding member and former Tesla AI director has joined Anthropic to work on pre-training, building a group that uses Claude to accelerate pretraining research. Eureka Labs, the education company he founded in 2024, is paused rather than closed.

May 2026
joined Anthropic
2015
OpenAI founding member
5 yrs
Director of AI at Tesla
Free
his courses remain

Key Takeaways

  • 01

    Karpathy joined Anthropic's pre-training team in May 2026, forming a group that uses Claude to speed up pretraining research itself.

  • 02

    He was a founding member of OpenAI, then Director of AI at Tesla, then returned to OpenAI, before founding Eureka Labs in 2024.

  • 03

    Eureka Labs is paused, not shut down. His teaching work, including Neural Networks: Zero to Hero and nanoGPT, remains free and widely used.

  • 04

    The move puts one of the field's best-known educators inside a frontier lab's core training effort.

What Happened

Andrej Karpathy joined Anthropic's pre-training team in May 2026, reported by TechCrunch on the 19th. The role is not a research-advisor seat: he is building a group that uses Claude to accelerate pretraining research, which means using the model to help improve how the next model is trained.

Why It Matters

Karpathy is unusual in having been at the centre of three distinct eras: a founding member of OpenAI in 2015, Director of AI at Tesla through the Autopilot years, then back at OpenAI, then out on his own with Eureka Labs. Very few people have built at that level and then taught it publicly for free.

Pre-training is also the least glamorous and most consequential part of the stack. It is where capability is actually set, before any fine-tuning or product work.

What He Leaves Behind

Eureka Labs is paused rather than wound up. His teaching material stays where it was and stays free: Neural Networks: Zero to Hero, Let's build GPT from scratch, nanoGPT, minGPT, and nanochat. Those remain among the most recommended free resources for learning how a language model actually works, and several of them anchor this site's own learning roadmap.

This article summarizes contemporaneous reporting.

AnthropicAndrej Karpathypre-trainingClaudeAI education
View Original Source