Hallucination

A model producing fluent, confident, wrong output — the inherent result of training for plausible next tokens rather than verified facts. Mitigated by retrieval, citations, validation, and honest UX; never fully eliminated.