A recent pre-print study by Swiss researchers, published on February 1st, highlights significant limitations in current generative artificial intelligence (AI) models, including ChatGPT, Gemini, and Claude. The study, which involved a benchmark test of 950 questions across legal, research, medical, and programming domains, found that AI models struggle with complex, multi-part queries, exhibiting a notable increase in "hallucinations" (inaccurate or false answers) in follow-up questions. The best-performing AI in the study had a global hallucination rate of 30%, while the least performant exceeded 70%. This research challenges the claims of AI companies and reignites the debate on whether AIs should admit when they don't know an answer rather than providing potentially incorrect information.