Six articles · September 2026

About this series

Andrej Karpathy gives his lectures away for free. He was a founding member of OpenAI, led the computer vision team behind Tesla's Autopilot, and now spends part of his time teaching. His talks are long, generous and unusually clear, and they are aimed at people with some technical background.

We watched just over six hours of them: the three-and-a-half-hour Deep Dive into LLMs like ChatGPT, his 2025 keynote Software Is Changing (Again), his two-hour Let's build GPT lecture, and a short talk from his Tesla years. We took notes, went back over them, and wrote out what we understood in plain words, for people who are not engineers and who want to understand what they are using, simply and practically.

Every idea in these six articles is his, except for the fourth, which is our own afternoon of asking, read through his pictures. Every article says so in its first paragraph and in a credit block at the end, and every picture carries his name. What is ours is the order, the drawings, the small experiments you can try yourself, and any mistake in the retelling. Where we say something wrongly, the fault is ours, not his.

Where he says it better than we can, we quote him, briefly, and tell you where to find the original. Watch the originals. They are better than any retelling.

Sources

• Deep Dive into LLMs like ChatGPT, Andrej Karpathy, YouTube, February 2025 (3 h 31 min).
• Software Is Changing (Again), Andrej Karpathy, Y Combinator AI Startup School, June 17 2025 (40 min).
• Let's build GPT: from scratch, in code, spelled out, Andrej Karpathy, YouTube, January 2023 (1 h 56 min).
• Multi-Task Learning in the Wilderness, Andrej Karpathy, ICML workshop, June 2019 (15 min).
• For article 6: the QLoRA paper by Tim Dettmers, Artidoro Pagnoni, Ari Holtzman and Luke Zettlemoyer (2023); the LoRA paper by Edward J. Hu and colleagues at Microsoft (2021); the DoRA paper by Shih-Yang Liu and colleagues at NVIDIA (2024).

Read the series

1. 01What you are actually talking toArticle 1 of 6 · a plain-language tour of the thing behind the chat box · after Andrej KarpathyEverything in this article is Andrej Karpathy's teaching, retold in plain words.
2. 02How it is built, told as schoolArticle 2 of 6 · from a copy of the internet to something that answers you · after Andrej KarpathyThis article retells the middle of Andrej Karpathy's Deep Dive into LLMs like ChatGPT, the part where he walks through how one of these models is actually made.
3. 03Why "show your work" is physics, not mannersArticle 3 of 6 · the one rule that makes a model smarter, and why it works · after Andrej KarpathyThis article retells one section of Andrej Karpathy's Deep Dive into LLMs like ChatGPT, the one he titles "models need tokens to think".
4. 04The test you can run yourselfArticle 4 of 6 · one football match, thirteen assistants, and what an afternoon of asking taught us · after Andrej KarpathyThe first three articles retold Andrej Karpathy's ideas.
5. 05How to work with one without getting burnedArticle 5 of 6 · the slider, the loop, the leash, and the suit · after Andrej KarpathyThe first three articles were about what the model is.
6. 06What fine-tuning changes, and what QLoRA isArticle 6 of 6 · the stage you can afford to run yourself · after Andrej Karpathy, and after Dettmers, Pagnoni, Holtzman and ZettlemoyerThe first four articles were Andrej Karpathy's.