Free courses · alignment & safety · start from zero
Beginner-friendly courses on how AI systems work — and how we try to align them.
I'm an independent researcher building a twelve-month behavioural study of an AI agent. These are the materials I needed when I started — built properly, and given away.
Free courses, explainers and reference material on how AI systems work and on alignment and safety. Written for complete beginners and for anyone who wants a direction into the field. No maths background assumed, no jargon left unexplained, no charge — ever. New material goes up often.
Get an email when there's something new
New modules go up often. Leave an address and I'll let you know — or just write to me. You're literally talking to me, not my team. I don't have a team.
No spam, no selling your address, ever. Unsubscribe in one click.
Courses
Each course is a stack of parts. Each part is a handful of modules. You can stop at the end of any part and have something whole.
/
//
Each module is its own designed piece — running analogy, diagrams, just-in-time maths only where it earns its place, and the wrong turns in the margin. It opens here, without leaving the site.
/ /
Before you start
How to actually learn this
About 40 minutes. It works far better if you do it in this order — and the order isn't a preference, it's what the evidence on learning says.
Tip: you don't have to type. Use voice-to-text and just say your answers out loud — it's faster, and saying it aloud is its own kind of recall.
Re-reading feels like learning and mostly isn't. What builds a memory is pulling the thing back out of your head — closing the page and saying it in your own words, badly, from memory. It feels worse and works better. That is why the notes come before the next section, and why the gate is closed-book.
A field of deep grass
There is a field of deep grass between you and an idea. The walking makes the path.
↓ Click it. Fourteen times.
Crossing 0 — deep grass, no path. You've read it once; you've never gone back for it.
The path was not there and then found. The walking made it. Re-reading is studying a map of the field: you can stare at it all afternoon, feel thoroughly familiar with the terrain, and have made no path at all.
Every analogy breaks somewhere, and this one breaks here: a real path only needs traffic. A memory needs the gaps. One hard crossing after a week beats five easy ones today — which is exactly why the modules never ask you to cram.
And why the visuals sit next to the writing rather than replacing it: the same idea in two forms gives you two ways back to it later. Not because anyone is a "visual learner" — that idea has fifty years of failed evidence behind it — but because two encodings of one thing beat one.
You almost certainly already have it
On Windows press Win + H. On a Mac, turn on Dictation in System Settings and press the mic key. On a phone, tap the microphone on your keyboard. Google Docs has Voice Typing built into the Tools menu.
Those cover almost everyone, and they cost nothing. Start there.
If you want something more accurate
Two paid tools are worth knowing about: Willow Voice and Wispr Flow. Both are paid, we are not affiliated with either, we make no money from them, and none of these are affiliate links. Neither is necessary — the free options above are genuinely fine for everything on this site.
Talk it through
You press one button and paste. That is the whole job. The button copies a prompt with the entire module already inside it — you don't have to read the prompt, and the assistant doesn't have to look anything up or open any files. It arrives knowing exactly the material you just read.
It will talk the module through with you, answer the questions you wrote down, and push back when you're vague. Stay there as long as you want. It will not test you — that is a separate button.
Then close the book
Only when you feel ready, press the second one. That copies the closed-book gate: it carries the module's own questions — not questions the assistant invented — and tells it to score you strictly and to explain what you misunderstood rather than just giving you the answer.
It does not matter which one
Claude, ChatGPT, Gemini, an open-source model you run yourself. Free tiers are fine.
Downloads this module as a folder you keep. Unzip it wherever you like and you get:
The module page works with no internet at all — everything it needs travels with it. Download a later module and its folder drops in beside this one, so the collection assembles itself as you go.
This screen shows every time. It is one click to pass, and it is the difference between reading a module and learning one.
Quiz Me
Pick what you've covered and how many questions you want. This page chooses them — weighted towards whatever you haven't been asked in longest — and copies them to your clipboard together with the module text they came from, so the assistant can't invent an answer. Paste it into any AI.
Step 1 — which course
Step 2 — which modules
Step 3 — how many questions
Step 4 — your questions
Pick some modules above.
Which questions you've been asked, and when, is remembered in this browser only — no account, nothing sent anywhere. Clear your browser or switch device and it simply starts picking at random again.
/ / /
How to learn this
Materials are the easy half. Most people teaching themselves lose to method, not to difficulty. This is the loop I run, every day.
Read once, properly
One module. No skimming ahead, no second tab. If your attention has gone, stop — a distracted pass costs you the module twice.
Write your own notes
In your words, not the module's. If you can only copy the sentence, you don't have it yet. The rewriting is the learning.
Get tested cold
Close everything. Answer from memory. Recall builds the memory — re-reading only builds the feeling of knowing.
“Ask me open questions on this, one at a time. Don't show me the answer. Tell me what I missed.”
Argue with the gaps
Where you were wrong, find out why the wrong version felt right. That reason is the actual lesson.
Come back later
Days later, cold, unannounced. Spacing is uncomfortable and it is the single largest effect in the learning literature.
Explain it out loud
Without notes. The place you stall is the place you don't understand. Go back to that place only.
Why this page exists: I trained in psychology before I came to this field. Retrieval practice, spacing and self-explanation are among the best-evidenced findings in the science of learning — and almost nobody self-teaching actually uses them. Sources and a longer write-up to follow here.
Papers, explained
These are other researchers' papers, retold in plain language so you can follow the argument without the jargon. None of them are mine — I have published no papers. Alongside them: a glossary, and a reading path that says why each item is on it.
Contact
If something here is wrong, unclear, or missing — or you just want to ask a question about any of it — write to me. I read everything.
I'm especially glad to hear from people working through this on their own, with no course and no cohort. That's who the site is for.
muhammad@muhammadhanzla.com
Every message reaches me directly. No automated replies.
Research enquiries, funding and commissioned work
I am available for paid research work. Study design, behavioural coding and analysis, and long-run observation of AI systems — as a commission, a contract, or a collaboration.
I am also actively seeking grant funding for the twelve-month behavioural study and the work that follows it. If you fund independent alignment and safety research, I would like to hear from you.
Professional work is carried on through Voyager Enterprise LLC, a Connecticut company. Write to the address above — it reaches me directly.
Get an email when there's something new
New modules and new material go up often. Leave an address and I'll write to you.
You're literally talking to me, not my team. I don't have a team. Reply to anything and it comes straight to me — questions, ideas, or telling me a module is wrong.
No spam, no selling your address to anyone, ever. Unsubscribe in one click.
Found a problem? Want something added?
Pick what kind of message this is, write it, and it goes straight to me.
Pick a type above to start.
About
I'm Muhammad Hanzla, an independent alignment & safety researcher. I'm running a twelve-month behavioural study of an AI agent, and I write these courses alongside the work.
I came to this from psychology, which matters more than it sounds like it should: the study I'm building is a long-run behavioural observation of an AI agent, and behavioural method is what I was actually trained in.
Everything here is free and always will be. The field needs more people in it than it currently has, and the bottleneck for most of them isn't intelligence — it's not knowing where to start.
The research is my work, not a hobby. The twelve-month study is the main thing I do. Alongside it I take on commissioned and contract research — study design, behavioural coding and analysis, and long-run observation of AI systems — and I am actively seeking grant funding to support the ongoing work. The courses on this site are free and stay free. The research is what I am paid for.
This work is carried on through Voyager Enterprise LLC. If you fund research, or you want a study designed or analysed, get in touch.