Why do AI models sometimes 'hallucinate' or create incorrect information?

AI models sometimes hallucinate or make up things because they're trying to guess what comes next, just like when you finish someone else's sentence before they’re done talking.

Imagine you're playing a game where you have to fill in the blanks of a story. You read the first few words: “The cat jumped over the ___.” You might say “mouse” or “dog,” depending on what makes sense to you. But sometimes, especially if the story is confusing or incomplete, you might guess something that doesn’t actually fit, like “spaceship.”

AI models work in a similar way. They look at words they’ve seen before and try to predict what comes next. If there are a lot of choices, or if the sentence is tricky, they might pick one that feels right, even if it’s not accurate.

Think of an AI model like a really smart kid who's been reading a lot of books. When you ask them to continue a story, they try to remember which words usually go together. But just like the smart kid might guess wrong sometimes, the AI can hallucinate, that means it makes up something that seems correct but isn’t.

That’s why sometimes AI says things that aren’t true, it's just doing its best guess!