Andrej Karpathy
Pretraining research team, Anthropic
- LLMs
- AI Education
- Agents
Editorial bio
Andrej Karpathy is a Slovak-Canadian AI researcher whose public teaching has shaped a generation of ML engineers directly. BSc in Computer Science & Physics from the University of Toronto, MSc from UBC, PhD from Stanford under Fei-Fei Li (thesis on connecting images and natural language). Founding research scientist at OpenAI (2015–2017), then Director of AI / Autopilot Vision at Tesla (2017–2022), where he led the shift to end-to-end neural networks for Full Self-Driving. A brief return to OpenAI (2023–2024) followed.
In July 2024 he launched Eureka Labs, an AI-native education company, and in 2026 joined Anthropic's pretraining research team. Alongside the day jobs, he's been the primary instructor behind Stanford's CS231n and the creator of "Neural Networks: Zero to Hero," nanoGPT, micrograd and minGPT – the minimal, readable artifacts that are, for a lot of practitioners, the way they actually learned deep learning. He coined or popularised "Software 2.0" and "vibe coding."
Worth following if you want to see what AI education looks like when it's taught by someone who still writes the code.
Key facts
- based in
- Bay Area
- current
- Pretraining research team, Anthropic (since 2026)
- prior
- Founder, Eureka Labs (2024–2026) · Research Scientist, OpenAI (founding team, 2015–2017; again 2023–2024) · Director of AI / Autopilot Vision, Tesla (2017–2022)
- education
- PhD, Stanford (advisor: Fei-Fei Li) · MSc, UBC · BSc (CS & Physics), University of Toronto
- recognition
- TIME100 AI · MIT Innovators Under 35
- focus
- LLMs · agentic systems · neural networks from scratch · AI education
Featured work
Anthropic
Current role.
Eureka Labs
AI-native education company.
"Neural Networks: Zero to Hero"
Line-by-line course on building LLMs from scratch; definitional for a generation of ML engineers.
nanoGPT · micrograd · minGPT
Minimal, readable reference implementations that doubled as teaching artifacts.
CS231n at Stanford (primary instructor)
The course that trained much of the modern computer-vision community.
"Software 2.0" & "vibe coding"
Coined / popularised two of the field's working terms.