FA
Top-down deep learning: you train a working model in lesson one and learn the theory as you need it.
Learn
51 courses across 6 tracks, and 42 of them cost nothing. The best AI education on the internet is free — Stanford, MIT, Google, Microsoft, Hugging Face and Andrej Karpathy all publish theirs openly. Every link on this page is fetched and checked each time the site is built, so you will not land on a dead page.
How the models actually work. Start here if you want to understand rather than just use.
16 courses 🧩Turning a model into working software: APIs, retrieval, evaluation, deployment.
13 courses 🤖Systems that plan, call tools, and finish multi-step work without a human in the loop.
5 courses 💬The cheapest skill with the highest return, and the one most people skip.
4 courses 💼Non-technical courses for people who need to use this at work on Monday.
5 courses 🎓Subscription libraries worth knowing about when you want depth or a certificate.
8 coursesIf you only do one thing, do one of these.
Top-down deep learning: you train a working model in lesson one and learn the theory as you need it.
Builds a neural network, then a GPT, from scratch in plain Python. The clearest explanation of backpropagation anywhere.
Google's own internal ML primer, with interactive exercises and no setup required.
A national-scale AI literacy course with no maths prerequisite. Finland taught this to its entire population.
An interactive textbook where every concept comes with runnable code in PyTorch, TensorFlow and JAX.
The course that trained a generation of computer vision researchers. Notes and assignments are public.
Stanford's natural language processing course, updated each year to cover current model architectures.
MIT's week-long bootcamp, published free every year with lectures, slides and labs.
A free certification track covering TensorFlow and the core ML workflow end to end.
Short, practical micro-courses that run in the browser. The intro to ML track takes an afternoon.
OpenAI's own educational resource for reinforcement learning, with reference implementations.
A 12-week, 24-lesson curriculum covering symbolic AI, neural networks, vision and NLP.
Where this comes from: Courses are selected by hand, then every URL is fetched at build time and dropped if it does not resolve. Last refreshed 22 July 2026. Nothing on this page is a paid placement.