AI, plainly

How AI chatbots work, one small piece at a time

Start with lesson 1 See all 15 lessons
Clarence Boey on the Brooklyn Bridge, with the Lower Manhattan skyline behind him

Why I made this

Clarence Boey · Undergraduate student ยท AI compliance and software engineering intern, NYC

Hi there! I'm Clarence. I'm an undergraduate computer science student from Singapore, and right now I'm interning in New York City, building AI compliance systems to help companies understand their agentic workflows and how AI agents are using their data internally.

I made this site because I kept seeing the same gap after speaking to my friends, family and the people I meet. They use AI chatbots every day, for work, for school and for everyday questions, but almost nobody has been told how they actually work. What is out there tends to be either a list of tricks or an explanation full of technical words, and neither one helps you decide when to trust an answer.

So I tried to write the explanations I wish I had known earlier when I began using these tools. I started by breaking down bigger concepts into smaller more digestible ones for others to understand. No technical background needed, and no fancy maths needed. If you can use a chatbot, you can follow every lesson here!

What this is trying to do to you

Two things, pulling in opposite directions. Make you more willing to try things, and be better at judging what comes back. Those turn out to be separable: in a survey of several hundred knowledge workers, more confidence in the AI went with less critical thinking, while more confidence in oneself went with more of it (Lee et al., 2025). Most beginner-friendly AI content turns up the first dial and leaves the second alone. This site is built to move both.

All 15 lessons

They build on each other, so the order is the recommended one. Each opens by naming the wrong version of the idea, because that is usually the version in your head, and it is easier to replace something you can see. 14 of them carry an interactive 3D diagram; every diagram has a written equivalent that teaches the same point on its own. Lessons marked try it have a small hands-on tool as well. None of them is a test.

What it is

Part 1 · Lessons 1–3 · about 6 min

A chatbot is a text predictor in a friendly window. It builds each answer one piece at a time, and it rolls the dice as it goes.

  1. What you are actually looking at3DSomething in there is thinking about my question.
  2. Where the words come from3DIt looks the answer up.
  3. Why it never gives quite the same answer twice3DWhat it said is what it thinks.

Why it gets things wrong

Part 2 · Lessons 4–5 · about 5 min

Invented facts and confident tone come from the same process as the right answers. Nothing in the writing tells you which one you got.

  1. Why it invents things that sound right3Dtry itIt made a mistake.
  2. Confidence is a writing style3Dtry itIt sounded sure, so it was probably right.

What it holds and what it can see

Part 3 · Lessons 6–8 · about 7 min

What it learned in training, what is in this chat, and what it can look up are three different things, and each one has limits.

  1. How much it holds at once3DIt can see everything we have said.
  2. Three kinds of memory3DIt remembers me, so it must be learning from me.
  3. What it can and cannot see right now3DIt is connected to the internet.

What it is good at, and where it is not

Part 4 · Lessons 9–10 · about 5 min

Sounding like it knows is not the same as knowing, and being good at one task tells you little about the task next to it.

  1. Writing like someone who knowstry itIt understands what I am asking.
  2. Good at that, bad at this3Dtry itIt is either good or it isn't.

Using it well

Part 5 · Lessons 11–15 · about 13 min

How to ask, what to make of a refusal, checks that take seconds, what to keep out of the chat box, and why trying things is safe.

  1. Asking as specification, not instruction3Dtry itIf I explain clearly, it will do what I want.
  2. When it declines or hedges3DIt is refusing because someone decided I should not know this.
  3. Verification that costs almost nothing3Dtry itI should check everything it tells me.
  4. What happens to what you type3Dtry itWhatever I type goes into the AI.
  5. Nothing here breaks3Dtry itI should probably learn this properly before I start.

How to use it

Read a lesson, then do the thing at the bottom of it before moving on. Each one takes under a minute and needs no preparation. The lessons are what change your model of the tool; the small actions are what make it stick. If you only have ten minutes, read lessons two, three and five.

When you are done, the toolkit collects the practical tools and a one-page card you can print and keep beside you.

There are no accounts, no tracking, and nothing to sign up for. Nothing you type into the diagrams or tools leaves your browser: the numbers in them are written into the page, not fetched from a live model.

This site is research and educational material, not advice about any particular tool or situation. Where a claim rests on thin evidence or on the author's judgement rather than published findings, the lesson says so in a line like this one.