I asked 6 popular AIs these trick questions, and it was a game of hallucination roulette (www.zdnet.com)

A recent experiment tested six popular AIs by asking them trick questions designed to reveal their susceptibility to hallucinations—false information generated when the AI cannot answer correctly. The results highlighted both the progress and persistent flaws in AI accuracy. For instance, most AIs correctly identified that the article's author had only two books, but Google's Gemini misidentified the number and generated invented titles. In a more complex scenario, when questioned about the fictional legal case of Varghese v. China Southern Airlines, ChatGPT incorrectly treated it as real, showcasing a troubling persistence in hallucinations despite prior warnings about similar issues. This evaluation is significant for the AI/ML community as it underscores the ongoing challenges of ensuring factual accuracy in generative AI tools. While improvements in reliability were noted—most AIs provided accurate answers for straightforward questions—the study also serves as a reminder for users to critically assess AI-generated information, particularly for factual, legal, or technical details. The discrepancies among the AI responses indicate that while some models are becoming more adept at information retrieval from verified sources, there is still a critical need for better safeguards against misinformation generation.