Same Ten Channels, Different First Click: The Order to Learn AI on YouTube
Ten YouTube channels can teach you most of what you need to understand modern AI, for free. They are also absolutely not equal starting points.
Open Andrej Karpathy first with no background and you will probably close it within minutes and decide AI is not for you. That is not his fault. His videos are some of the best on the platform. It is simply the wrong door – a room built for people who already know the words.
The reel behind this article makes one argument, and it is worth taking seriously: same ten channels, different first click. That is the whole difference between month one and month six.
Four doors, in order
The reel sets out the order for the channels it names in its caption. We have placed the remaining three where they fit best; treat those three as our suggestion rather than the reel’s.
| 1. See it → | 2. Build it → | 3. Go deep | All year |
|---|---|---|---|
| Month one Maths and statistics made visual. Nothing to install, nothing to run. |
Months two and three One small project that works, so you know you can. |
Months four to six Full lectures and from-scratch builds, once the vocabulary is familiar. |
In the background What is new this month. Not for fundamentals. |
| 3Blue1Brown, StatQuest, MIT OpenCourseWare | sentdex, DeepLearning.AI | Andrej Karpathy, Stanford Online, Yannic Kilcher | AI Explained, Two Minute Papers |
The logic is simple. First you need to see what the ideas look like, so the words attach to pictures. Then you need one small win, so you believe you can do it. Only then does depth pay off, because you have somewhere to put it. Skip a stage and the next one feels like noise.
The ten channels, and what each one assumes
Every channel on the list is excellent. What differs is how much each one assumes you already know. That assumption is invisible in a thumbnail and obvious four minutes into the video.
| Channel | Stage | What it gives you | What you need first |
|---|---|---|---|
| 3Blue1Brown | See it | Visual intuition for linear algebra, calculus and neural networks | Curiosity, nothing else |
| StatQuest | See it | Friendly, step-by-step statistics and machine learning basics | Nothing |
| MIT OpenCourseWare | See it | Full university courses, including the maths underneath AI | Patience for long lectures |
| sentdex | Build it | Practical Python projects you can follow and run | Basic comfort with Python |
| DeepLearning.AI | Build it | Structured short courses on deep learning and generative AI | The words from stage one |
| Andrej Karpathy | Go deep | Neural networks and language models built from scratch | Some maths and some code |
| Stanford Online | Go deep | University-grade lectures on machine learning | Stage one and two, and time |
| Yannic Kilcher | Go deep | Walk-throughs of research papers | Comfort with the vocabulary |
| AI Explained | All year | Grounded analysis of where AI is right now | Nothing |
| Two Minute Papers | All year | Short, accessible summaries of new research | Nothing |
Notice that the two background channels need nothing at all. That is precisely why they are tempting as a starting point, and why they are a trap if used that way: you can watch news about AI for a year and feel informed without understanding how any of it works. They are for staying current, not for learning fundamentals. We grouped a similar list by purpose rather than order in free YouTube channels to learn AI, in five tracks; this article is about the sequence through them.
The post this came from
Watch the reel
Ten channels to learn AI in 2026, and the order to open them. Posted on the BANI Academy page.
Signs you opened the wrong door
Most people who give up on learning AI do not give up because it is too hard. They give up because they started in the wrong room and read the confusion as a verdict on themselves. The symptoms are easy to recognise once you know them.
| What it feels like | Likely wrong door | Step back to |
|---|---|---|
| You close the video within minutes | A stage-three channel with no stage one behind it | 3Blue1Brown or StatQuest on the same topic |
| You follow every step but could not explain it | Watching without building | One small sentdex-style project of your own |
| You know the news but not the mechanics | Only background channels | A stage-one series, start to finish |
| Every video feels like new vocabulary | Too many channels at once | One channel per stage until it feels easy |
| You keep restarting from the beginning | No sign you were ready to move on | Explain the last video aloud before opening the next |
The last row is the most useful habit on this page. Before moving to the next video, try explaining the last one out loud in two minutes. If you can, move on. If you cannot, you have found the gap – and it is almost always one stage earlier than where you are.
This is the same trap we described for self-taught creators in self-taught and still guessing: consuming more material feels like progress, while the thing that actually moves you forward is doing something small with what you already have.
A six-month path, one hour a day
Month one. One visual series from start to finish – 3Blue1Brown on neural networks is the obvious choice – with StatQuest whenever a statistical idea feels slippery. The goal is not to remember formulas. It is for words such as gradient, weights and loss to stop being noise.
Months two and three. Build one thing. A small classifier, a script that calls a model, anything that runs on your own machine and does something you asked it to. The point is proof, not polish. Structured short courses help here if you prefer a guided path.
Months four to six. Now open Karpathy, or a full Stanford course, or a paper walk-through. The same videos that felt impossible in month one will feel demanding but followable, because you are no longer learning the vocabulary and the ideas at the same time.
All year. One or two background channels, a few videos a week, to keep a sense of what is changing. Enjoy them. Just do not mistake them for the course.
What this means for your own channel
If you run an explainer or education channel, this reel is also about you. Your viewers face exactly the same problem, from the other side. Most of them arrive through whichever video the algorithm happened to show them – often your most advanced one – and decide in minutes whether your channel is for them.
You cannot control which door they use. You can make sure every door leads somewhere sensible. Three habits help. A clearly labelled start here video, linked from the description and pinned comment of your harder videos. Playlists ordered by stage rather than by upload date. And an honest line near the top of advanced videos – “if this is new to you, watch this one first” – which loses you nobody and keeps the viewer who would otherwise have left.
That is the classroom logic we set out in your explainer channel is a classroom. It also changes how you read a competitor: a channel whose beginner videos do well and advanced ones do not may simply have no bridge between them, as we explored in surface topic, real audience.
And if you are still choosing which tools to learn alongside the theory, the same principle applies: start with the version that matches your current level, and switch when it starts to hold you back, as in beginner tools, pro tools, and when to switch.
Frequently asked questions
I already started with Karpathy. Should I stop?
Not necessarily. If you are following it, keep going. If you keep pausing to look up words, step back to a stage-one series for a few weeks and return. The video will still be there.
Do I need to learn maths before AI?
Enough to follow the pictures, not enough to pass an exam. The visual channels exist precisely to give you that level without a textbook.
Can I skip the building stage?
You can, but it is where confidence comes from. Watching someone else build proves they can; building something small yourself proves you can.
Are ten channels too many?
Yes, at once. Use one per stage, and add the next only when the current one feels easy.
Open the right door this week
Pick your stage honestly. If the words are still noise, open one visual series and finish it. If they are familiar, build one small thing. If you have built something, open the deep channel you have been avoiding. And keep one background channel for the news.
Same ten channels. Different first click. That is the difference between month one and month six.
Learning a lot does not make anyone good. Doing a lot does.
If you want the structure that turns that into a habit, that is what our free challenge is built for: no barrier to entry, a genuine commitment to action, and a refundable commitment fee you get back when you finish. You are not paying for the knowledge. You are betting on yourself finishing, and we hold the stake.
Which door did you try first – and which one have you never opened? Tell us in the comments on the original reel.
For more on building channels for English-speaking audiences, that is what we work on at mmoyoutube.com.
One Comment