Hallucination (AI)
When an AI confidently makes up facts that aren't true.
LLMs are trained to produce plausible text, not necessarily true. If you ask about an obscure person or specific date, the model can generate an answer that sounds reasonable but is entirely fabricated.
Mitigation strategies: RAG (only answer from source texts), citation requirements in the prompt, fine-tuning on fact-focused datasets, or human review.