Hallucination Means a Plausible False Answer
An AI can generate something that sounds real even when the information is false. People often call this a hallucination.
It may invent a name, date, study, quotation, website, court case, or explanation.
Why Does This Happen?
A language model is built to generate a useful continuation from learned patterns. Producing plausible language and verifying truth are not exactly the same job.
Confidence Is Not Accuracy
An AI can use confident language even when it is wrong. More detail can make a false answer feel more believable, but detail is not evidence.
Grounding Gives the Model Evidence
One way to reduce hallucinations is to give the AI reliable information to work from. This is called grounding. Search, trusted documents, databases, and specialized tools can all provide evidence.
Tools Can Do Jobs Models Should Not Guess At
For exact arithmetic, use a calculator. For account balances, use the trusted database. For current news, use current sources. Strong systems do not force the language model to do every job itself.
The Big Idea
AI generation and fact-checking are different jobs. Good architecture can add search, tools, verification, or human review when accuracy matters.