Language models, such as those used in chatbots and virtual assistants, sometimes produce fictional or made-up information when answering questions. OpenAI's research found that this 'hallucination' is caused by the model's tendency to fill in gaps in its knowledge with plausible but incorrect information. Improved evaluation methods can help identify these issues and make language models more reliable. --- Why it matters: This matters because it highlights a fundamental limitation of current language models, which can have significant implications for applications where accuracy is crucial, such as customer service or medical diagnosis. Source: https://openai.com/index/why-language-models-hallucinate