Andrej Karpathy
Andrej Karpathy is an AI researcher and educator, a founding member of OpenAI and the former Director of AI at Tesla. He is best known for the Stanford CS231n deep learning course, his public teaching videos, and open-source projects such as nanoGPT and nanochat. He founded the AI education company Eureka Labs in 2024, and in May 2026 announced that he joined Anthropic.
Background and notable work
- Academic background: he earned a BSc in computer science and physics from the University of Toronto, an MSc from the University of British Columbia, and a PhD in computer science from Stanford University in 2015, advised by Fei-Fei Li; his thesis work combined images and natural language. At Stanford he authored and taught the deep learning course CS231n.
- OpenAI: he joined as a founding member in 2015 as a research scientist and left in 2017.
- Tesla: from June 2017 he served as Director of AI, working on Autopilot Vision; he announced his departure in July 2022.
- Second stint at OpenAI: he announced his return in February 2023, where, by his own account, he built a new team working on midtraining and synthetic data generation; he left in February 2024.
- Eureka Labs: an AI education company he announced in July 2024, with LLM101n as its first course project.
- Open-source projects: nanoGPT (a minimal codebase for training medium-sized GPTs), minGPT, llm.c and nanochat. nanochat, released in October 2025, is a full end-to-end pipeline for training and serving a ChatGPT-like model on a single GPU node; its README states that around 100 dollars of training cost yields a GPT-2-capability model.
- Teaching and writing: the “Zero to Hero” technical series and general-audience LLM explainer videos on YouTube; blog posts such as “Software 2.0” and “A Recipe for Training Neural Networks”. In February 2025 he coined the term “vibe coding” for building software by prompting AI with natural language.
The above timeline follows his homepage and the English Wikipedia entry. On May 19, 2026, he announced on X that he joined Anthropic, and the company said he would lead a pretraining research team.
Mentioned on the show
Karpathy did not take part in the recording; the following is the hosts’ discussion. In Weekly #004’s chapter on founder-driven companies and frontier models, Xiangyang Qiaomu said that “around the start of the year, a whole wave of X influencers — for example AK — joined Anthropic” (see that line in the Chinese transcript). “AK” refers to Karpathy, consistent with his publicly announced move to Anthropic in May 2026.
Frequently asked questions
Who is Andrej Karpathy?
A founding member of OpenAI and former Director of AI at Tesla, the lead author and instructor of the CS231n deep learning course, author of nanoGPT and nanochat, and founder of the AI education company Eureka Labs (2024).
What does Andrej Karpathy do at Anthropic?
On May 19, 2026, he announced that he joined Anthropic, and the company said he would lead a pretraining research team. (Editorial addition, based on the public announcement cited in his Wikipedia entry.)
Did AK join Anthropic?
Yes. “AK” refers to Karpathy (a separate researcher using the display name AK as @_akhaliq on X is unrelated): on May 19, 2026 he announced on X that he joined Anthropic, and the company said he would lead a pretraining research team (see his homepage and the public announcement cited in his Wikipedia entry). In Weekly #004, Xiangyang Qiaomu also mentioned that “a whole wave of X influencers — for example AK — joined Anthropic”, a host’s relay of that move (see that line in the Chinese transcript).
Who coined the term “vibe coding”?
Karpathy, in February 2025, describing a way of building software where natural-language prompts drive AI-generated code and the human mainly steers and reviews the result.
What are nanoGPT and nanochat?
His open-source training codebases: nanoGPT is a minimal implementation for training and finetuning medium-sized GPTs; nanochat, released in October 2025, covers tokenization, pretraining, finetuning, evaluation, reinforcement learning and inference end to end, with the goal of training a usable ChatGPT for under 1,000 dollars.