AI新闻可靠性报告亮红灯，45%回答存重大错误，Gemini成重灾区

Core Insights - A large-scale study across 18 countries and 14 languages reveals systemic flaws in mainstream AI chatbots, with nearly half of the responses containing significant issues, posing a serious threat to the quality of public information retrieval [1][4] Group 1: Study Findings - The research, conducted by 22 public service media organizations, tested major AI platforms like ChatGPT, Copilot, Gemini, and Perplexity with 3,000 evaluations, finding that 45% of responses contained at least one major error, while 81% had minor issues [4] - The two main areas of concern were the lack of information sources and insufficient accuracy, affecting 31% and 30% of responses respectively [4] - The study's lead researcher warned that these errors are systemic, cross-border, and multilingual, leading to public distrust in information sources [4] Group 2: Platform Performance - Google Gemini had the highest error rate at 76%, more than double that of other products, indicating significant reliability issues [4] - Despite some improvements in models compared to six months ago, the error rates remain high, particularly in handling breaking news, complex timelines, and fact-based questions requiring viewpoint differentiation [4] Group 3: AI Adoption Trends - In contrast to the reliability issues, the adoption of AI for news consumption has surged, with user numbers doubling within a year, especially among young people aged 18-24 in Argentina and the United States [4] - Alarmingly, 48% of young users rely on AI to simplify news understanding, highlighting a growing dependency on these tools despite their shortcomings [4]