Some of the best AI education in the world is free on YouTube. World-class researchers and teachers explain everything from the intuition behind neural networks to building a language model from scratch. Here are the channels and courses we recommend, organized by level.

Start here: intuition

3Blue1Brown — Neural networks series

Grant Sanderson’s 3Blue1Brown is famous for beautiful animated math explanations. Its neural networks series explains what a neural network is, how gradient descent works, and what backpropagation actually does — later extended with chapters on large language models, Transformers and attention. Best for: anyone who wants deep intuition without heavy prerequisites.

StatQuest with Josh Starmer

Clear, step-by-step explanations of statistics and machine learning concepts — from logistic regression and decision trees to neural networks and Transformers — with memorable humor. Best for: beginners who want to understand the building blocks.

Build things: hands-on courses

Andrej Karpathy — Neural Networks: Zero to Hero

Andrej Karpathy, a founding member of OpenAI and former director of AI at Tesla, created a series that builds neural networks from scratch in Python: starting with a tiny automatic differentiation engine (“micrograd”), then character-level language models, and finally a GPT-style Transformer. He also published accessible talks such as an introduction to large language models, and a long video on how models like ChatGPT are built. Best for: programmers who want to understand how language models really work.

fast.ai — Practical Deep Learning

Jeremy Howard’s course takes a top-down approach: build working models first, understand theory later. Best for: coders who learn by doing.

University lectures

  • MIT 6.S191 Introduction to Deep Learning: a compact, updated-yearly introduction with lectures and labs.
  • Stanford CS231n: convolutional networks and computer vision, with influential lecture recordings.
  • Stanford CS224n: natural language processing with deep learning.
  • Stanford CS229: classic machine learning, more mathematical.

Best for: structured learners who want academic depth.

Keeping up with research

Two Minute Papers

Károly Zsolnai-Fehér presents exciting research results — especially in graphics, simulation and generative AI — in short, enthusiastic videos. Best for: staying inspired by what’s possible.

Yannic Kilcher

In-depth paper walkthroughs and discussions of machine learning research. Best for: technical viewers who want to understand papers.

How to learn effectively from videos

  1. Code along instead of just watching.
  2. Pause and predict what comes next.
  3. Rebuild without the video after finishing a section.
  4. Take notes in your own words.
  5. Combine with reading — documentation, papers and books.
  6. Build a project applying what you learned.

A suggested path

  1. 3Blue1Brown’s neural networks series for intuition.
  2. StatQuest for core machine learning concepts as needed.
  3. Karpathy’s Zero to Hero or fast.ai for hands-on building.
  4. A university course for depth in your area of interest.
  5. Research channels to stay current.

Building a study plan around videos

A realistic plan for someone studying part-time:

  1. Weeks 1–2: watch 3Blue1Brown’s neural networks chapters, taking notes in your own words.
  2. Weeks 3–6: work through the first lessons of a hands-on course, coding along and completing exercises.
  3. Weeks 7–10: follow Karpathy’s series to build a small language model, rebuilding each step without the video.
  4. Weeks 11–12: start a personal project applying what you learned, using documentation and forums when you get stuck.

Throughout, keep a learning log of concepts, questions and links. Revisiting it consolidates knowledge and shows your progress.

Avoiding tutorial hell

Many learners watch tutorial after tutorial without building anything independently. The fix is simple but uncomfortable: after each course section, build something small on your own, without following a video. Struggling through problems is where most learning happens.

What to watch next

  • New editions of university courses covering the latest models.
  • Video courses on AI agents and AI engineering.
  • Interactive and AI-tutored learning platforms.
  • Creators explaining reasoning models and multimodal systems.

Key takeaways

  • 3Blue1Brown builds deep visual intuition for neural networks and Transformers.
  • StatQuest explains machine learning building blocks clearly.
  • Karpathy’s Zero to Hero builds a GPT-style model from scratch.
  • MIT and Stanford lectures provide structured academic depth.
  • Code along and build projects to learn effectively from videos.

Frequently asked questions

Can I learn AI only from YouTube?

You can learn a great deal, especially combined with practice and projects. Structured courses and reading help fill gaps.

What should I watch first?

Start with 3Blue1Brown’s neural networks series for intuition, then move to hands-on courses.

Do I need math for these videos?

Intuition-focused videos need little; hands-on and university courses benefit from basic linear algebra, calculus and probability.