We rely on AI more and more for information, but what happens when AI encounters conflicting facts? Imagine asking a chatbot a simple question like "Who is George Washington?" and getting bombarded with details about both the first U.S. President *and* a lesser-known inventor and jazz musician, all with the same name. This isn't a hypothetical scenario. It highlights a critical challenge facing today's AI: knowledge conflicts. New research explores this exact problem by introducing 'WhoQA,' a dataset designed to test how Large Language Models (LLMs) handle conflicting information. Researchers discovered that even subtle conflicts significantly impact AI accuracy. When presented with multiple sources mentioning different George Washingtons, LLMs often falter, sometimes prioritizing popular figures or even ignoring the context entirely. Interestingly, they seem *less* sensitive to more obvious conflicts, suggesting a complex relationship between the amount of conflicting data and AI’s ability to process it. Why does this matter? Because these conflicts can lead to misinformation and biased responses, particularly in retrieval-augmented generation (RAG) systems, where AI retrieves information from external sources to answer questions. The WhoQA dataset uses real Wikipedia entries to create these conflict scenarios, making it a practical test for real-world applications. The study found that while some LLMs simply admit they can't answer, others make a potentially more damaging choice: picking one answer and ignoring the rest, leading to inaccurate and potentially biased results. Telling the LLMs explicitly about the *possibility* of conflicts helps, but it doesn't completely solve the problem. This research highlights the ongoing challenge of building truly reliable and trustworthy AI systems. Future research will likely explore fine-tuning methods to better equip LLMs to handle these unavoidable knowledge conflicts, paving the way for more robust and transparent AI that can navigate the complexities of the real world.