A study has found that a significant portion of references provided by ChatGPT, specifically model GPT-4, are fabricated or contain errors, with some reports suggesting that more than half or nearly two-thirds of AI-generated citations are inaccurate. This issue, known as "hallucination," has led to errors in various areas, including news, health advice, and research papers. The problem appears to be widespread, with AI assistants making mistakes that can slip past experts and enter the permanent record.