What is Hallucination?
When a model states something confidently that is false or unsupported by its data.
Formal Definition
A hallucination is model output that is fluent and confident but factually wrong or unsupported by the provided evidence. It arises because models predict plausible text, not verified truth. In a trading context, a hallucination is a model citing an earnings figure, price level, or trend that the supplied market data does not actually support, which the grounding rubric penalizes.
In Simple Terms
It is when an AI makes something up and says it with a straight face. The wording sounds right, but the fact is wrong or invented. In trading, that might be a made-up number that pushes a bad decision.
Example
A model claims a company beat earnings when the data shows a miss. The trade may look justified in the text, but it rests on a hallucinated fact.