The AI educator who builds from scratch
Andrej Karpathy
Profile
Andrej Karpathy is the rare researcher who can build a transformer from scratch on a livestream and make you feel like you could, too. A Stanford PhD under Fei-Fei Li, founding member of OpenAI, former Director of AI at Tesla where he led Autopilot vision for five years — his resume would be intimidating if he weren’t so generous with what he knows.
He left OpenAI a second time in February 2024 to focus on education, and in July 2024 founded Eureka Labs, an “AI-native” school aimed at teaching people how AI actually works. Before that, his YouTube channel quietly became one of the most important AI education resources on the internet. The “Neural Networks: Zero to Hero” series walks you from a scalar autograd engine (micrograd) up through a working GPT, one keystroke at a time. nanoGPT, llm.c, and makemore are the same philosophy as code: small, readable, hackable.
What sets Karpathy apart is his refusal to let abstraction hide the math. He’ll explain backpropagation by literally typing out the chain rule in a Jupyter notebook. He coined “Software 2.0” back in 2017 — the idea that neural network weights are a new kind of code — and more recently “vibe coding,” the now-ubiquitous term for letting an LLM drive your IDE while you ride shotgun. Both landed because he names the thing developers are already doing.
For anyone bridging theory and practice — a student who knows the math but hasn’t shipped, or a veteran engineer who ships daily but wants to finally understand attention — Karpathy is the bridge. He’s not selling a course or a framework. He’s just showing his work.
Key Articles & Papers
The Unreasonable Effectiveness of Recurrent Neural Networks Software 2.0 A Recipe for Training Neural Networks Deep Reinforcement Learning: Pong from Pixels Deep Visual-Semantic Alignments for Generating Image Descriptions What I learned from looking at 200 machine learning tools Yes you should understand backpropVideos
Spotify Podcasts
