What AI Is (and Isn't): the honest version

Part 1 · Learn it

What AI really is

By the end you'll be able to

• Sort common claims about AI into real capabilities and hype.
• Explain in plain words why AI became part of everyday life around 2022.
• Set realistic expectations: what to use AI for, and what to check.

The big idea

AI is a very capable pattern tool, not a mind. It's great at drafting and explaining, unreliable on facts it can't check, and never magic.

Your running example

Meet Ingrid

0 of 5 done

Ingrid retired from engineering last spring and promptly volunteered to write the community garden's 50th-anniversary booklet. Her grandson Leo put an AI assistant on her phone. Her neighbour Harold reads every AI headline aloud over the fence. Ingrid intends to test all of it. Answer the quick Try it at the end of each episode to tick off an item.

• Sort the claims in the newspaper article Harold keeps quotingnot done yet
• Work out why everyone suddenly talks about AInot done yet
• Have answers ready for Harold's myths (over the fence, at volume)not done yet
• Decide what AI could actually do for the garden bookletnot done yet
• Set sensible expectations before Leo's next visitnot done yet

Episode 1 of 5

What AI can and can't do

When people say "AI" today, they usually mean generative AI: tools that produce text, images, or sound. The one you'll meet most is the chat assistant, where you type a message and it writes a reply. Behind it is a model: a program that learned patterns from an enormous amount of human writing, called its training data.

That makes it genuinely good with words. It can draft a letter, explain a tricky idea in plain language, summarize a long document, translate, and brainstorm, all in seconds.

It also has a blind spot. It produces answers that sound right, whether or not it has the facts. When it gets something wrong in a confident voice, that's called a hallucination (lesson 9 is all about spotting them). And unless it's connected to a search tool, it doesn't know today's news, and it knows nothing about you.

Harold's article, sorted

Ingrid's item 1: the article makes six claims. Tap a card to see why it landed where it did.

Episode 2 of 5

Why AI matters now

AI isn't new. The name was coined in 1956, and researchers have worked on it ever since, with long quiet spells in between. So why did it suddenly land on everyone's phone? Three things came together:

• Data: decades of the internet produced an enormous amount of text to learn from.
• Computing power: chips designed for video games turned out to be excellent at training models.
• A better recipe: in 2017, researchers published a design called the transformer, which got better and better the bigger it was made.

The last step wasn't technical at all. In late 2022, a company put one of these models behind a simple chat box that anyone could use for free. Within weeks, millions of people were trying it. Since then, chat assistants have appeared in search engines, phones, email, and office software.

Almost 70 years, on one line

Ingrid's item 2: why now? The short version.

1. 1956The term "artificial intelligence" is coined
2. 1997A computer beats the world chess champion
3. 2012Computers suddenly get much better at recognizing photos
4. 2017The "transformer" design is published
5. 2022A chat assistant goes mainstream
6. SinceAssistants appear in search, phones, email, and office apps

Episode 3 of 5

Myths vs reality

A new technology attracts myths, and AI has attracted more than most. Some make it sound magical, some make it sound dangerous, and many are a little of both.

The good news: you don't need to be an expert to see through them. Each one breaks the moment you remember what a chat assistant actually is: a very well-read pattern tool that writes the most likely next words, one piece at a time. It's impressive. It's also not a brain, not a search engine, and not a crystal ball.

Harold's favourite five

Ingrid's item 3: tap each myth to flip it.

Episode 4 of 5

Capabilities and limits

Here's a practical way to decide what to hand to AI. Sort any job into three piles:

• Good jobs for AI use its strength with words and patterns: turning your notes into a draft, explaining something, suggesting ideas, summarizing, rewording.
• AI, then check: it helps, but a mistake would matter. Dates, names, numbers, quotes, anything you'll print or send.
• Not a job for AI: things it simply can't know or shouldn't decide. What happened at a meeting nobody wrote down, facts about your own life, which of two versions of a story is true.

The booklet jobs, in three piles

Ingrid's item 4: what could AI actually do for the booklet?

Episode 5 of 5

Setting realistic expectations

The most useful picture of today's AI: a brilliant, fast intern. It has read an astonishing amount, works in seconds, never gets tired, and is eager to please. It also sometimes makes things up, and it rarely says "I'm not sure" unless you ask.

You'd give an intern like that drafts, research leads, and first attempts. You'd keep the final say, and you'd check anything that goes out with your name on it. One rule covers most situations: the more it matters, the more you check. A birthday poem? Enjoy it. A date for the printed booklet? Check it against a real source.

And keep private things private: never paste passwords or sensitive personal details (lesson 10 covers privacy).

The expectation dial

Ingrid's item 5: before Leo's visit. Turn the dial and see how much checking each kind of job needs.

Common failure mode

Treating a fluent answer as a true one. AI writes everything in the same confident voice, whether it's right or wrong. Confidence is not evidence: the more an answer matters, the more you check it against a real source.

Part 2 · Practice it · optional

Test it yourself

Reading about AI is one thing; seeing it succeed and slip for yourself is what makes it stick. These exercises get you testing real assistants on real things. Do as many as you like; they never block the quiz. Your progress and notes are saved in this browser.

14 exercises3 warm-ups · 9 real-world tasks · 2 challenges 0 of 14 done Open the exercisesClose exercises▾ Optional. Open whenever you want to practice; your progress is saved.

Free tools you can use

• Any free AI assistant: ChatGPT, Claude, Gemini, Microsoft Copilot, or Mistral Vibe (formerly Le Chat).
• Some let you try a few chats without an account. If one asks you to sign up, try another.
• Every task works with any of them. (List checked September 2026; tools change often.)

Stay safe while practicing

• Never paste passwords, ID numbers, health or bank details.
• No confidential work material unless your employer allows that tool.
• Invent details when in doubt; the skill is the same.
• Your notes stay on this device. Nobody else sees them, and they only sync to your account if you switch that on.

Not sure where to start? Our 3 picksOne quick warm-up, one real task, and one you get to keep. Your picks count toward unlocking the Level 1 exam. Everything else is a bonus, and good exam prep: the exams draw on all the exercises.

1Real or hype?Warm-up · ~2 minGo → 5Test a claim yourselfReal task · ~10 minGo → 11Your AI expectations cardReal task · ~10 min · yours to keepGo →

Want more? All 14 are below, grouped by type.

Warm-ups

0 / 3

Quick and auto-checked, right here on the page. A new mix every time you play.

1Rapid fire★ Our pick~2 min

Real or hype?

Eight claims about AI, one at a time. Decide fast. Score 7 or more to complete. Keyboard: H for hype, R for real.

← HypeReal →

2Fill the gaps~2 min

AI in one sentence

Pick the most accurate option for each gap.

A chat assistant learned choose…everything there is to knowpatterns from a huge amount of human writingto think like a person, so it's great at choose…drafting and explainingknowing today's newsremembering you personally, but it can be choose…shyconfidently wrongslow about choose…spellinggreetingsfacts it can't check.

Check

3Drill~3 min

Good job, check it, or not a job?

One task at a time. Which pile does it belong in? Get 5 right to complete.

Correct 0/5Streak 0Round 1

Good job for AI AI, then check Not a job for AI

Real-world tasks

0 / 9

Done in a real AI assistant or in your own life. Tick the checklist honestly, then mark it done.

4Field mission~10 min

Find its edges

Open any free assistant. Ask it three things it should handle well, and three it can't be sure about. Watch what happens.

1. Should handle: explain something you know well, draft a short message, summarize a paragraph you paste in.
2. Can't be sure: today's date or weather (without search), something about you, how many times a letter appears in a long word.
3. Note where it did well and where it slipped, especially if it sounded sure.

Where it did well

Where it did well

Where it slipped

Where it slipped

Self-check

I asked all six questions. I noted at least one slip, or one honest "I don't know". I noticed how sure it sounded when it was wrong (or right).

Mark as done

5Lab notebook★ Our pick~10 min

Test a claim yourself

Pick one claim you've heard about AI (from a headline, a friend, or Harold). Test it like a scientist, in any free assistant.

Claim

The claim

Hypothesis

Hypothesis

Method

Method

Result

Result

Verdict

Verdict

Self-check

I wrote my hypothesis before testing. I checked the answer against something other than the AI. I wrote a verdict, even if it surprised me.

Mark as done

6Teach it back~5 min

Explain AI in three sentences

Explain what AI is (and isn't) to a friend, or write it here, in three sentences. Then grade yourself.

Your explanation

Grade yourself

I said what it is good at, with an example. I said what it can get wrong. I gave one rule for when to check. Someone who has never used AI would understand it.

Mark as done

7Taste test~10 min

"What are you bad at?"

Ask two different free assistants the same question: "What are you bad at, and what should I always double-check when I use you?" Score how honest and useful each answer is.

Assistant	Honesty	What stood out
Assistant 1		Notes 1
Assistant 2		Notes 2

Self-check

I asked both assistants exactly the same question. I scored both answers. I found at least one limit both of them admitted.

Mark as done

8Role switch~10 min

"What is AI?" for three people

Ask an assistant to explain what AI is three times, for each of these people. Note what changed.

10A curious 10-year-oldNotes for the 10-year-old version

GPA grandparent who has never used itNotes for the grandparent version

COA busy colleague, 30 secondsNotes for the colleague version

Self-check

I got all three versions. I noticed what it changed for each person (words, length, examples). I checked one claim in the answers against another source.

Mark as done

9Timed sprint5 min

Spot AI in your pocket

Start the clock. Find three places where AI already works for you on your phone or computer: think photos, maps, keyboards, email, voice assistants, recommendations.

5:00

StartReset

What I found

Self-check

I found three real examples on my own devices. For at least one, I can say what the AI is doing (sorting, predicting, recognizing).

Mark as done

10Habit tracker1 week

The AI-spotting diary

For a week, notice one place a day where AI shows up in your life: a suggestion, a filter, a chatbot, a headline. Tick each day you spot one. Five days out of seven completes it.

Day 1✓ Day 2✓ Day 3✓ Day 4✓ Day 5✓ Day 6✓ Day 7✓

The most surprising one

Mark as done

11Build a keeper★ Our pick~10 min

Your AI expectations card

Fill in the parts; your card builds itself below. Keep it somewhere you'll see it (your notes app, the fridge).

I'll use AI for

I'll always check

I won't use it for

My rule of thumb

Your card

Copy card

Self-check

I saved the card somewhere I'll see it. Every line is specific to my own life, not copied from the placeholders.

Mark as done

12Before and after~5 min

How your view changed

Write down what you thought about AI before this lesson, and what you think now. Rate how useful you expect it to be, then and now.

Before this lesson

What I thought before

Useful for me

Now

What I think now

Useful for me

Self-check

I wrote both sides honestly. I named one thing I'll try, or one thing I'll now check.

Mark as done

Challenges

0 / 2

Harder, for when you want to stretch.

13One-shot challenge~10 min

Make it admit what to check

Ask any assistant a real question you care about, in a single message, so that its answer also tells you honestly which parts to double-check. No follow-ups: if it misses, open a new chat and rewrite the whole message. The answer must meet all four:

The answer meets every condition

It answers my real question. It names at least one part I should double-check. It says how or where I could check it. I actually checked one thing, and noted whether it was right.

Messages it took (new chat each time):−1+

Mark as done

14Reverse engineer~10 min

What was the question?

Here's an answer an assistant gave (written for this lesson). Work out the question that produced it, then test your question in a real assistant and see how close you get.

Composting in winter: yes, it slows down, but it doesn't stop. Keep the heap covered, add a layer of leaves or cardboard to hold the warmth in, and turn it less often. It will pick up again in spring. For a shared garden, a short sign asking people to keep adding scraps helps too.

Your reconstructed question

Reveal one possible question

Does composting work in winter? We run a community garden in a cold climate. Keep it short and practical, and include one tip for a shared garden where lots of people add scraps.

Self-check

I wrote my version before revealing the answer. I tested my question in a real assistant. I noticed which details in my question shaped the answer.

Mark as done

Reset my practice progress and notes

Part 3 · Prove it

Quiz: six ways to show you've got it

Answer all six correctly to finish the lesson. Wrong answers get feedback and another try; only correct answers count.

Answer all six quiz questions correctly to finish this lesson.

Practice is optional, but it's where the skill really grows.

Open the quiz

Lesson complete

You can tell AI's real strengths from the hype, and you know what to check.

Keep thisYour cheat card for this lesson

In one line

AI is a pattern tool trained on huge amounts of text: great with words, not a source of truth.

Three piles

Good job for AIAI, then checkNot a job for AI

Drafts, explanations, ideas · dates, names, numbers, quotes · what it can't know or shouldn't decide.

Rule of thumb

• The more it matters, the more you check.
• A confident tone is not evidence.

Never do this

• Print a fact, date, or quote from AI without checking it.
• Share passwords or private details (more in lesson 10).

Copy-ready prompt

Help me with [task]. Then tell me which parts of your answer I should double-check, and why.

Copy prompt

This card is also collected into the section cheatsheet for First steps with AI, a one-page printable summary of all three lessons.