How does generative AI hallucinate information?

Generative AI sometimes makes up information because it tries to guess what comes next, like a kid trying to finish a story before it's done.

Imagine you're telling a story to your friend, and every time you say a word, they try to figure out what you’ll say next. Sometimes they get it right, “Once upon a time,” they might say, “there was a… castle!” But other times, they guess wrong, “Once upon a time,” they say, “there was a… robot!” That's kind of like how generative AI works.

Generative AI looks at what it already knows and tries to predict the next part. If it sees a lot of stories about dragons, it might think the next word is dragon. But if it doesn't know much about robots, it might say robot even when it's not right.

Sometimes AI gets confused because it’s trying too hard to make sense of everything. Imagine you're reading a book with big words and strange pictures, you might guess the story is about space ships, but it could really be about cookies! That’s how generative AI can hallucinate, it makes up things that sound right, but aren’t actually there.