How AI Models Work

A blank page becomes a paragraph, illustration, melody, or block of code in seconds. It looks like the system created something from an idea—but the process is closer to guided prediction than imagination. Generative AI builds new outputs one small step at a time. How can that produce useful creativity and convincing mistakes through the very same mechanism?

The answer includes a convincing name, an exact date, and even a source. There is only one problem: none of them are real. AI hallucinations are not strange acts of imagination. They happen when a model keeps producing likely-sounding text after reliable information runs out. Why doesn’t it simply stop and say, “I don’t know”?