Understanding AI assistants’ mistakes and limitations
AI produces the most likely answer, which is not necessarily the right one. It can be confidently wrong, be unaware of recent facts or refuse to answer a legitimate question.
Common mistakes
- Making up a fact or a source.
- Getting a calculation wrong.
- Not knowing about a recent event.
- Refusing to answer an ordinary question out of caution.
Learning to check information
- Compare it with a book or a reliable website.
- Ask an adult.
- Ask the question again in a different way.
What about Kidikia?
Kidikia is prompted to acknowledge when it is unsure, but it can make mistakes too.
Why AI makes mistakes
An AI model has learned to predict words from vast amounts of text. It does not “know” things in the human sense: it puts together the most likely answer. This leads to mistakes known as “hallucinations”, especially with numbers, dates, lesser-known people and current events.
Explaining it to a child
“AI is like a classmate who has read lots and lots of books but sometimes remembers things incorrectly. It sounds confident even when it is wrong. So for important things, we check.”
A short fact-checking exercise
- Ask: “How many legs does a spider have?”, then check in a book.
- Ask for a historical date and compare it with an encyclopedia.
- Look together for a deliberate mistake in a text written by AI.
This page is an automatic translation of the French version, which is the authoritative version. Written and reviewed by the Kidikia team · Last updated on October 1, 2026