Same mechanism, different outcomes. Fluency is not accuracy.
A confident storyteller who, when they don't remember the details, fills in the gaps with something that sounds right. They're not lying. They genuinely don't know the difference between what they remember and what they invented.
Use the arrows below, the dots above, or your keyboard arrow keys to move through the stages.
Both generated the same way. Photosynthesis matches reality. The researcher is fiction.
Fluency and accuracy are independent. Beautifully written and completely false can coexist.
Hallucination isn't a bug for the next version. It's the same mechanism that makes AI creative. You can tune the balance but can't eliminate it.
Knowing where hallucination is likely is your best defense.
Every response is a prediction, not a retrieval. The more specific the claim, the higher the risk. If it matters, check it.
Now you know why AI can produce confident fiction. But there's another limit -- why does AI seem to forget what you told it earlier in the conversation? →