The best free Karpathy AI courses, curated
Andrej Karpathy — founding member of OpenAI and former director of AI at Tesla — makes some of the clearest deep-learning teaching anywhere, and gives it away for free on YouTube with matching code on GitHub. Below is his work organized into a path: start by building neural networks from scratch, then go deep on how modern LLMs actually work.
These are curated links to videos and code published freely by Andrej Karpathy on YouTube and GitHub. All content belongs to Andrej Karpathy. AnybodyCanAI is not affiliated with, sponsored by, or endorsed by him. His name is used only to identify the source.
Neural Networks: Zero to Hero
Build neural nets from scratch, ending with a GPT
Andrej Karpathy · 2022–2023 · Video series
Summary
A hands-on, code-first series that builds neural networks from the ground up: first a tiny autograd engine (micrograd), then character-level language models (makemore), and finally a working GPT and its tokenizer — spelled out line by line.
The single best way to actually understand what happens inside a neural network and a transformer. Every lecture is a full-length video paired with runnable code, so you learn by building — no black boxes. If you do one thing on this hub, do this.
Schedule
- 1The spelled-out intro to neural networks and backpropagation: building micrograd
- 2The spelled-out intro to language modeling: building makemore
- 3Building makemore Part 2: MLP
- 4Building makemore Part 3: Activations & Gradients, BatchNorm
- 5Building makemore Part 4: Becoming a Backprop Ninja
- 6Building makemore Part 5: Building a WaveNet
- 7Let's build GPT: from scratch, in code, spelled out
- 8Let's build the GPT Tokenizer
Access
Every lecture is a free YouTube video with matching Jupyter notebooks in the nn-zero-to-hero repo. No signup. Basic Python and a little calculus help.
Materials & links
Source: Andrej Karpathy — all links open on the provider's own site.
Deep Dive into LLMs like ChatGPT
How modern LLMs are really built, end to end
Andrej Karpathy · 2025 · Video series
Summary
A comprehensive (3.5-hour) walkthrough of the full LLM stack: pretraining on internet text, tokenization, supervised fine-tuning, reinforcement learning from human feedback, and how inference and hallucination actually work.
The best single explainer of how a model like ChatGPT is built, for people who want the real mechanics without a math degree. Perfect after Zero to Hero, or on its own if you mainly want to understand rather than build.
Schedule
See the course site under Materials for the current schedule.
Access
A single free YouTube video (~3h31m). No signup.
Materials & links
Source: Andrej Karpathy — all links open on the provider's own site.
Intro to Large Language Models
The busy person's 1-hour intro
Andrej Karpathy · 2023 · Talk
Summary
A one-hour, general-audience introduction to large language models: what they are, how they are trained, where they are headed, and the emerging security issues around them.
The fastest way to get a genuinely accurate mental model of LLMs. No code, no prerequisites — ideal for founders, PMs and the merely curious before diving deeper.
Schedule
See the course site under Materials for the current schedule.
Access
A single free YouTube video (~1 hour). No signup.
Materials & links
Source: Andrej Karpathy — all links open on the provider's own site.
How I Use LLMs
A practical guide to using AI tools well
Andrej Karpathy · 2025 · Talk
Summary
A practical tour of how Karpathy actually uses LLMs day to day — model selection, tools, multimodal features, coding assistants and real workflow examples.
Most people underuse the tools they already have. This is a concrete, example-driven guide to getting far more out of everyday AI assistants — squarely on-mission for AnybodyCanAI.
Schedule
See the course site under Materials for the current schedule.
Access
A single free YouTube video. No signup.
Materials & links
Source: Andrej Karpathy — all links open on the provider's own site.