Autocomplete
Free · no sign-up needed
Free · hands-on education · no sign-up needed

Learn how to use AI through demonstrations

Eight short lessons inside a working replica of an AI. Learn what it can see, what it forgets, and where it goes wrong.

Try the first lesson → See the curriculum
autocomplete · what-is-an-llmLive · try it
Why it works

You learn the machine by sitting inside a replica of it.

Most “intro to AI” content is a wall of text or a talking head. Autocomplete is different: every lesson is a small interactive experience presented inside an interface that behaves like the real thing. The medium is the message.

  • 01

    Felt, not read

    You draw the conclusion yourself by watching the model respond — the “aha” lands because you caused it.

  • 02

    No background assumed

    Written for owner-managers, founders and teams who’ve used ChatGPT as a better Google and want to use it properly.

  • 03

    One idea at a time

    Each module builds on the last, starting from the keystone: an LLM predicts the next word, over and over.

The whole idea, in one sentence

The best way to learn something new is to  

Scroll — watch it guess the next word, and the next.

The curriculum

Eight modules, start to finish.

Begin anywhere — but modules one and two are where it clicks. Your progress is saved on this device automatically.

  1. 01 The foundation What is an LLM (Large Language Model) Run four steps and watch the model predict text one word at a time. The keystone idea everything else builds on: it isn't thinking, it's autocomplete. 6 min Completed →
  2. 02 Orientation Getting Around the Interface A guided tour of every control in a modern chat app — model picker, thinking effort, files, voice, projects, history and temporary chat — then three worked examples through the same mock UI. 8 min Completed →
  3. 03 What it can see Context — What the Model Can See Ask for a client email with no context and get “Dear [Client Name]”. Attach a client list, a product one-pager and a voice guide in the chat, ask again — and watch a generic reply become a specific one. 7 min Completed →
  4. 04 Working memory The Context Window A live token meter is the hero. Attach one novel to a 128k model and it answers; attach a second and the window overflows — the first book silently drops out of view; switch to a 1M model and both fit. 7 min Completed →
  5. 05 The payoff Projects — Set Context Once Create a five-a-side coaching project yourself — name it, add the instruction and sources — then run three brand-new chats — website, break-even, marketing — each auto-loading the same context. Fresh never means from scratch. 8 min Completed →
  6. 06 Automation Skills & Schedules Package a repeated procedure once so it runs the same way every time, fires with one line, and can be put on a schedule. Build a /morning-summary skill and set it for 8am. 8 min Completed →
  7. 07 Make things HTML — The Presentation Layer Describe a dashboard in plain English and watch the AI write HTML, then render a rich interactive artifact — KPI cards, charts and a data table. The whole course is AI-generated HTML. 7 min Completed →
  8. 08 Use it safely Risks & Issues Where these tools go wrong and how to stay in control — confident-but-wrong answers, privacy and security, and the habits that keep AI useful without trusting it blindly. 7 min Completed →
About

The name is the thesis.

These tools aren’t “intelligence” the way the marketing implies — they’re very sophisticated autocomplete machines. Every Autocomplete lesson is built to make that click, then build outward into the things you can genuinely do once you understand the mechanism. It’s the same approach Exponential Partners uses with client teams: start by understanding the tool, then ship the work that follows.

Start with lesson one →
Scripted by design

Every response is scripted. That’s the point.

There is no live AI model behind these lessons — and that’s a decision, not a shortcut. A scripted lesson lands the same way for every learner, every time: the trick question always tricks, the fix always fixes, and nothing hallucinates while it’s teaching you about hallucination. It also means there’s nothing to sign into, nothing to pay per answer, and a lesson can run anywhere — including embedded on someone else’s site.

Who’s behind this

Built by someone who does this for a living.

  • MIT AI Product Design
  • 200+ professionals trained
  • 20+ AI tools built

I’m Riz Pabani. I’ve spent 20 years in tech and business in London, the last few focused almost entirely on AI — building tools, testing models, and teaching people to actually use them. I did an AI Product Design programme at MIT and I’ve trained over 200 professionals.

I built Autocomplete because I kept having the same conversation: smart people, senior roles, using ChatGPT like a faster Google with no idea what else it could do. These are the exact lessons I run with client teams through Exponential Partners — made hands-on so the ideas land on their own. If it clicks for your team, that’s literally my day job.

Bring it to your team →
Riz Pabani, AI trainer, London
Riz Pabani · London
Questions

Fair questions, straight answers.

Do I need any technical background?
No. If you can use a browser, you can do this — everything is shown, not explained. You click, the machine responds, and the idea lands on its own.
I already use ChatGPT a bit — will I learn anything?
Almost certainly. Using it and understanding it are different things: if you can’t yet say why it makes things up, what a long chat does to the context window, or when to reach for a thinking model, the middle modules pay for themselves.
Which AI tool does this teach — ChatGPT, Claude, Gemini?
All of them and none of them. You learn inside a replica app built from the same controls every real tool shares, so you learn the mechanism once and carry it everywhere.
How long does it take?
About an hour. Eight modules of six to eight minutes each, self-contained, with your progress saved — so you can do one over coffee and pick the rest up later.
Are the AI responses real?
They’re real responses a model would give — captured and scripted so the teaching is deterministic. You’re watching the machine’s genuine behaviour, on rails.
Why not wire up a live model?
Because a live model is unpredictable in exactly the way a lesson can’t afford — the trick question has to trick every time. When you’re ready for the unpredictable version, you’ll have the mental model to handle it.
Is anything I type sent to an AI?
Nothing. There’s no live model behind these lessons, so nothing you type is sent anywhere, stored, or used to train anything. It’s also why the course is free — there’s no per-answer bill to pay.
Is it free? Do I need an account?
Free, and no account needed. Progress saves on your device; an optional sign-in makes it follow you across devices.
Why is it called Autocomplete?
Because that’s the secret. Underneath the marketing, the machine is predicting the next word, over and over — and once that clicks, everything else stops being mysterious.
Can I share this with my team or embed it?
Yes. Every lesson runs anywhere, including embedded on another site. If you want the whole thing run for a team, that’s my day job — get in touch.

Ready to see how it actually works?

Six minutes from now, “autocomplete” will stop being a metaphor and start being obvious.

Begin the course →

Send feedback

Found a bug or have a thought? Tell me — it comes straight to my inbox.