Hallucinations occur because an LLM is trained to produce fluent text, not necessarily true statements. Typical examples are fabricated sources, incorrect figures, or non-existent quotes. Techniques such as RAG, low temperature, and clear factual instructions reduce the risk considerably. Critical review of the outputs nonetheless remains essential, especially for factual or legal topics.
Hallucination
A hallucination is an AI-generated statement that sounds plausible but is factually wrong or fabricated. It arises because language models predict probabilities rather than look up facts.