AI Hallucination
When an AI model generates confident but factually incorrect, fabricated, or nonsensical information that has no basis in its training data or reality.
Why It Happens
LLMs are trained to produce plausible text, not truthful text. They interpolate patterns from training data without a grounded understanding of facts. When uncertain, they may generate convincing-sounding but fabricated details.
Mitigation
RAG (grounding in retrieved documents), tool use (fact-checking via search), chain-of-thought (explicit reasoning), fine-tuning on factual data, and constitutional AI methods that train models to express uncertainty.