Hallucination
When a language model states something false as if it were fact — a confident, fluent answer that simply isn't true. It happens because the model predicts plausible text, not verified truth.
Think of It Like This
Like a smooth talker who would rather sound sure than admit they don't know.
A model generates the most likely next token, and "likely" is not the same as "correct," so gaps in what it learned get filled with confident guesses. Fabricated citations, invented API methods, and wrong dates are all classic cases. Grounding the model in real sources at answer time — retrieval-augmented generation — is the most reliable way to cut hallucinations down, since the model quotes retrieved facts instead of improvising.