As @nprofile1q... points out, AI/LLMs don’t actually hallucinate. They confabulate. This is also called “making shit up”. For example, when a person with severe dementia tells you a very plausible story about how they arrived at the ER today, but five minutes later they tell a completely different tale, they are confabulating. Their brain knows they should have a story, so they just start filling gaps. LLMs do this. It is unpredictable, and extremely hard to detect in a written result.