Hallucination
A hallucination is when an AI system states something false as if it were fact — an invented figure, a policy that does not exist, a citation to a document never written — with exactly the same confidence it uses when it is right.
Also written: confabulationArabic: الهلوسة
Why the confidence is the dangerous part
An obviously wrong answer is harmless; someone catches it. The costly case is a plausible one — a number in the right range, a clause that sounds like your contract — because it passes review. The risk is not that the model is wrong, it is that being wrong looks identical to being right.
What actually reduces it
Grounding answers in retrieved documents and showing the sources. Letting the system say it does not know. Testing against a fixed set of questions with known answers, re-run whenever anything changes. Note that none of these are model choices — they are design choices around the model.
Questions people ask
No. It can be reduced a long way and made visible, which is a different and achievable goal. Any deployment premised on it never happening is designed wrong.
Related serviceAI for Business