Hallucination

Definition: When a model states something confidently that isn't true — not because it's lying, but because it always generates the most statistically plausible next words, and plausible isn't always the same as correct.

Why it matters: This is the single biggest reason AI output needs verification before it goes anywhere that matters — dates, citations, statistics, and legal or medical claims are exactly where plausible-sounding and actually-true are most likely to diverge.

Example: Asking for an academic source on a niche topic and receiving a real-sounding author, journal, and year for a paper that doesn't actually exist.