The Best AI YouTube Channels for Education
ZarifYouTube is where I learned most of what I know about AI. Books are slow. Courses are expensive. The right channels give you world-class teaching for free — if you know which ones to watch.
Updated September 16, 2026 — channel URLs, cadence, and reasons verified.
Last verified September 16, 2026. Recheck target: monthly. Download CSV (19 entries).
Three filters: does the host understand the material, or are they reading a generated script? Is the content evergreen or pure news churn? Production quality matters less than clarity.
Start with five
Fundamentals, internals, research awareness, and one practitioner who ships automations.
- 3Blue1Brown · Monthly-ish
Grant Sanderson's neural-network and transformer series is still the best visual introduction to the math.
- StatQuest with Josh Starmer · Weekly-ish
Slow, no-prerequisites explanations of statistics and ML. Use this when a paper assumes you already know boosting.
- Andrej Karpathy · Sporadic
The build-GPT and tokenizer walkthroughs explain how language models work through code. Useful when you want to implement the components yourself.
- Two Minute Papers · Two to three times a week
Short videos on one paper at a time, mostly graphics, robotics, and generative work. Stay broadly aware without reading every PDF.
- Nick Saraev · Weekly
Agency and automation builds from someone who has sold the work. Workflows, clients, and what actually ships.
The directory
Use fundamentals channels when you do not yet have the math in your bones. Use news channels for a week in review, then open the paper. Use agent and n8n channels only if that is the work in front of you.
Scroll horizontally to compare the columns.
| Name | Role | Why it is here | Cadence |
|---|---|---|---|
| 3Blue1Brown | ML fundamentals | Grant Sanderson's neural-network and transformer series is still the best visual introduction to the math. | Monthly-ish |
| StatQuest with Josh Starmer | ML fundamentals | Slow, no-prerequisites explanations of statistics and ML. Use this when a paper assumes you already know boosting. | Weekly-ish |
| Andrej Karpathy | LLM internals | The build-GPT and tokenizer walkthroughs explain how language models work through code. Useful when you want to implement the components yourself. | Sporadic |
| Yannic Kilcher | Paper walkthroughs | Reads the paper slowly, with intuition. Best for intermediate viewers who want DeepSeek or Llama explained from the PDF. | Weekly-ish |
| AI Explained | News analysis | Sourced, adult takes on model releases. The antidote to hype YouTube. | Two to three times a week |
| Two Minute Papers | Research awareness | Short videos on one paper at a time, mostly graphics, robotics, and generative work. Stay broadly aware without reading every PDF. | Two to three times a week |
| Matt Wolfe | Tool roundups | Weekly news rundowns plus FutureTools. Use for a 20-minute synthesis of what shipped, then open the original. | Two to three times a week |
| Matthew Berman | Hands-on demos | Tests new models the day they drop. Less theory, more what it actually does in a browser. | Daily |
| Wes Roth | News analysis | Synthesizes frontier news with a what-does-this-mean frame. Higher hype than AI Explained; pair them. | Daily |
| Nick Saraev | Agents and automation | Agency and automation builds from someone who has sold the work. Workflows, clients, and what actually ships. | Weekly |
| Cole Medin | Agents and automation | Agent frameworks, MCP, and orchestration with code on screen. Creator of Archon. | Two to three times a week |
| Liam Ottley | Agents and automation | The business mechanics of selling AI services: sales, pricing, positioning. Skip if you only want architecture. | Weekly |
| Nate Herk | n8n | Tutorials combine n8n workflows with AI tools. Useful for seeing the connections and setup steps before building your own workflow. | Two to three times a week |
| Fireship | Dev news | Five-minute, code-bearing takes the day a model drops. Highest density of any AI-adjacent dev channel. | Weekly |
| James Briggs | RAG and embeddings | Vector databases, RAG, and applied LLM engineering with working code, assuming you want to ship. | Weekly |
| Sam Witteveen | Framework tutorials | New tools and Colab notebooks within a week of a framework or model drop. | Weekly |
| Riley Brown | Vibe coding | Non-engineer builds with Claude Code, Cursor, and Replit. Useful for where product-building is going, not for ML internals. | Two to three times a week |
| David Shapiro | Forecasting | Opinionated takes on AI safety, AGI, and post-labor economics. Engage; do not treat as calibrated timelines. | Two to three times a week |
| AI Engineer | Conference talks | Talks from AI Engineer Summit and World's Fair. The closest thing to a proceedings YouTube can be. | Around events |
If you only have 30 minutes a week for AI YouTube news, watch one AI Explained video and one Matthew Berman demo. That combo gives you the analyst view and the user view of whatever shipped.
Conference talks live on the AI Engineer channel; they pair with AI communities and the X accounts that clip them.
See the change log below for updates to this directory.
Change log
Moved onto the directory template with start-with-five, per-entry cadence, and a CSV download of channel URLs.