Why language models hallucinate

OpenAI has published new research explaining the underlying causes of language model hallucinations. The study demonstrates how better evaluation methods can improve AI systems' reliability, honesty, and safety performance.