Large language models (LLMs) are impressive, but they sometimes "hallucinate," meaning they confidently generate incorrect information. One tricky type of hallucination is when LLMs contradict established facts. Researchers have developed a clever tool called "Drowzee" to detect these fact-conflicting hallucinations. Drowzee works by building a vast knowledge base from sources like Wikipedia. It then uses logical reasoning rules to create complex questions and their correct answers. For example, if Drowzee knows Bob Dylan won the Nobel Prize in Literature and that Haruki Murakami hasn't, it can generate a question like, "Did Murakami and Dylan ever win the same award?" The correct answer, of course, is no. Drowzee then presents these questions to LLMs. To check if the LLM truly understands, Drowzee doesn't just look for a simple "yes" or "no." It analyzes the LLM's reasoning process, comparing its logic to the known facts. This helps identify when an LLM gets the right answer for the wrong reasons or uses incorrect information. The results are revealing: LLMs struggle with questions involving time, unfamiliar information, and complex logic. Drowzee's automated approach is a significant step toward making LLMs more reliable and trustworthy. It highlights the need for ongoing research into LLM hallucinations and offers a promising path toward mitigating these issues, paving the way for more robust and dependable AI systems in the future.