Hallucination
Also known as · confabulation
When a model produces fluent output that is factually wrong or made up.
A hallucination is when a model states something false with the same confidence and fluency as something true — inventing a citation, a statistic, a quote, or an API that doesn't exist. It happens because an LLM generates statistically plausible text; plausibility and truth usually overlap, but not always.
Hallucination is the single biggest reliability risk in deploying LLMs, especially in high-stakes domains like law, medicine, and finance. The model has no built-in sense of when it's guessing versus recalling.
Mitigations include grounding the model in real sources (RAG), asking it to cite and quote, lowering temperature for factual tasks, and keeping a human in the loop for consequential decisions. The honest framing: treat fluent output as a draft to verify, not an authority.