Hallucination

Hallucination is the failure mode in which a language model states something false with the same fluency and confidence it uses for correct answers. It happens because the model is trained to produce plausible continuations of text, not to check facts. When the training data is thin on a topic, or the prompt asks for a specific citation, date, or number the model does not hold, it fills the gap with a well-formed guess. Fabricated academic references, invented API methods, and confident but wrong legal citations are typical; lawyers have been sanctioned for filing briefs containing case names a chatbot made up. Mitigations reduce the rate without eliminating it. Retrieval-augmented generation grounds answers in retrieved documents, tool use replaces recall with lookup, and asking the model to cite sources makes errors easier to catch. Structured verification and human review remain necessary for anything published. The dangerous property is that hallucinations look exactly like correct output, so confidence in the wording carries no information about accuracy. Treating a model as a reasoning surface over supplied evidence, rather than as a knowledge store, is the practical response.