A hallucination is AI-generated content that appears to answer a question but is false, unsupported, or incoherent. A fluent response is not necessarily an accurate one. section 5.12 section 5.8
Different answers to the same question
Asked “Who is Akli Adjaoute?” in fresh sessions, ChatGPT produced different false biographies. It attributed an unrelated company, an award, and a PhD to him. The fifth attempt gave an accurate answer. Repeating a prompt did not make the earlier claims reliable. section 5.3.2
ChatGPT generates words in response to input; predicting plausible word combinations does not guarantee factual accuracy. False or biased input data can also influence LLM output. These limitations matter when generated answers are used to seek knowledge online or teach students: unchecked claims can displace testable sources. Check important factual claims against such sources rather than trusting an answer because it sounds convincing. section 5.3.2 section 5.10 section 5.8
