How Large Language Models are Designed to Hallucinate