Hallucination: fluent falsehood, not a typo

An AI hallucination is when the model generates a statement that reads fluently and plausibly but is factually wrong, unsupported, or entirely made up. Unlike a simple typo or grammar mistake, the error is in the content itself — for example, inventing a court case citation that looks real or claiming a historical event happened in the wrong year. The key contrast is between surface confidence and actual accuracy: the model does not know it is lying.

To spot a hallucination in exam questions, look for overly precise numbers, dates, or names that sound correct but are not common knowledge. If a generated answer includes a specific statistic or a quote from a person, treat it as a claim to verify rather than a fact. Another trick: ask whether the detail is something the model could plausibly have learned from its training data — obscure or invented specifics are red flags. Never assume that fluent language equals truth.

A compact way to remember this: confidence is not evidence. When you see a detailed AI output, mentally append the phrase "unless verified" to every specific claim. This turns a passive recall point into an active check — if you cannot quickly confirm the detail, treat it as a hallucination until proven otherwise.