“Extrinsic hallucination in large language models occurs when the model's output is fabricated and not grounded by its pre-training data or world knowledge.”