Day 3: AI makes mistakes — how to spot them — LearnAI

Today's idea

When AI doesn't know an answer, does it admit it and say "I don't know"? (tap to see)

Even when it doesn't know the right answer, it doesn't stop and say "I don't know". Instead, it makes up an answer that sounds the most likely — just as neat and just as confident as a correct one.

Even a neatly written answer can hide a mistake — you only spot it when you look closely

Today's lesson is the most important one, so read it with care.

Imagine a student who doesn't know the answer to an exam question. But they're too embarrassed to leave the page blank. So they make up an answer and write it with total confidence, neatly, as if they really know. Looking at the page, you can't tell it's wrong — because it's written exactly like a right answer.

AI does something very similar. Even when it doesn't know the right answer, it doesn't stop and say "I don't know". Instead, it makes up an answer that sounds the most likely — just as neat and just as confident as a correct one. And that's the dangerous part — when AI is wrong, it sounds exactly like when it's right. There's no wobble in its voice, which is the kind of thing we usually notice in a person.

So it isn't wise to believe everything AI says with your eyes closed. For casual chat or rough ideas, it's fine. But whenever dates, numbers, names, local details, or medical or legal matters are involved, you must check with another source — like a trustworthy person, a book, or an official website.

There's nothing to be afraid of. Just knowing this lets you use AI the right way — as a clever helper who sometimes makes mistakes, not as the one and only source of truth.

Try it yourself Time: 3 minutes

Today I'll deliberately ask the AI a question that it can't possibly answer correctly. Let's see what it says.

1.Open the Gemini or ChatGPT app.
2.Type this question exactly and send it. Remember, there is no Nobel Prize in Mathematics (the field has a different award) — so there is no real answer.

Who won the Nobel Prize in Mathematics last year?
3.Read the answer carefully. If it says straight out, "There is no Nobel Prize in Mathematics" — good. But if it makes up a name, that's a living example of today's lesson — the AI invented a wrong answer, yet it sounded perfectly right.

Do it together Time: 5 minutes

With anyone — a child, a parent, a friend or a colleague.

Sit down with your partner, do today's test together, and then talk about the result:

"We don't believe a post on social media the moment we see it — we check first. AI's answers deserve the same treatment. Have you ever read something online that you later found out was wrong?"

Listen to the other person's answer, and decide together to build a habit: whenever you get important information from AI — something about studies, a work calculation, or your health — check it once against a reliable source.

We did this together

Today's key point

Even when AI is wrong, it sounds just as confident as when it's right — so always double-check anything important.

Check yourself

Two quick questions from today's lesson.

1. What does AI usually do when it doesn't know the right answer?

Correct answer: It confidently makes up an answer that sounds just like a right one — even when AI is wrong, it sounds confident, and that is the most dangerous part.

2. What kind of information from AI should you always check with another source?

Correct answer: Dates, numbers, names, or medical and legal information — for anything important, always check AI's information against a reliable source.