What is a "hallucination" in AI models?

A hallucination is when an AI makes up something that isn’t true, just like a kid might say they saw a unicorn at the park, even though there were only dogs running around.

Imagine you're telling your favorite story to a friend. If you get excited and add details that weren’t in the original tale, like saying the dragon had wings made of candy, that’s kind of like a hallucination. The AI is trying to tell a good story, but it gets carried away and adds things that aren’t real.

How It Happens

Think of an AI model as a super smart robot who has read many books before. When you ask it a question, it tries to answer by picking bits from all the stories it knows. But sometimes it mixes up parts of different stories, just like when you mix up your socks in the laundry and wear one from yesterday with today's shirt.

Why It Matters

Sometimes these made-up parts are fun, like if the robot tells a story about a cat who flew to the moon. Other times, they might be confusing, like if it says 2 + 2 equals 5 just because it remembered a different math problem. But that’s okay! The AI is learning and growing, just like you do when you make mistakes while telling stories.

Take the quiz →

Examples

  1. An AI says the moon is made of cheese, but it's not true.
  2. A chatbot tells you a famous person said something they never did.
  3. An AI creates a fake recipe for a cake that doesn't exist.

Ask a question

See also

Discussion

Recent activity