Andrej Karpathy
Andrej Karpathy
If you want to actually understand how AI works — not just use it, but build it from scratch and know what every line does — Andrej Karpathy is your person. He’s that rare combination of world-class researcher and gifted teacher who makes hard things feel approachable.
His YouTube series “Neural Networks: Zero to Hero” is, in my opinion, the single best resource for learning AI from first principles. No hand-waving. No magic. You build everything yourself, and by the end, you’ve made a GPT.
The Career
OpenAI founding member — Was there from the start, helped build the systems that became GPT.
Tesla Director of AI (2017–2022) — Led the computer vision team for Autopilot. Real-world, safety-critical AI at massive scale.
Eureka Labs (2024) — Founded his own AI education company. Thesis: AI can provide 1-on-1 tutoring that adapts to each learner. Human-crafted curriculum + AI delivery.
Why He’s Special for Learners
Karpathy’s teaching philosophy is “build it from scratch.” His projects are legendary in the ML community:
| Project | What you learn |
|---|---|
| micrograd | Backpropagation in ~100 lines of Python. The whole engine. |
| makemore | Character-level language models — bigrams to Transformers |
| nanoGPT | A minimal GPT implementation you can actually run and understand |
| minbpe | Tokenisation from scratch — how text becomes numbers |
| “Let’s build GPT” | 2-hour video building a Transformer from nothing |
The progression is brilliant: each project builds on the last, and by the end you understand how LLMs work at a level most AI engineers don’t.
The Vibe
What I appreciate most is the honesty in his teaching. He’ll say “this is messy” or “here’s why this is confusing” instead of pretending everything is elegant. It feels like learning from a colleague, not a lecturer.
“The hottest programming language is English.”
That quote captures something important: as AI gets more capable, the skill of communicating clearly with it (see Prompt Engineering) becomes as valuable as traditional coding.
Start Here
If you’re new to his content, this is the order I’d suggest:
- “Intro to Large Language Models” (1 hour overview talk)
- micrograd (understand backpropagation)
- “Let’s build GPT” (build a Transformer)
- Then explore makemore, minbpe at your own pace
Go Deeper
- Neural Networks — Start here if Karpathy’s content is too fast
- Transformers — The architecture his “Let’s build GPT” constructs
- How LLMs Work — The conceptual overview to pair with his implementations
- Learning Resources — Other great educators and resources
- Training & Fine-Tuning — The process his tutorials teach you