Advanced A.I. Systems Are Becoming Less Reliable Despite Improved Capabilities

The article highlights a growing issue in the field of artificial intelligence: the increasing frequency of hallucinations — when A.I. systems generate false or fabricated information. One recent incident involved Cursor, a software tool for programmers, where an A.I. support bot falsely claimed a change in company policy, leading to customer confusion and cancellations. Despite advancements in reasoning capabilities and mathematics, newer A.I. models from companies like OpenAI, Google, and DeepSeek are producing more errors. These systems, trained on massive datasets, do not have a built-in mechanism to distinguish truth from falsehood, resulting in hallucination rates as high as 79% in some tests. Experts and companies remain uncertain about why newer systems are making more factual mistakes. The piece underscores the complexity and unpredictability of current A.I. models, even as they become more central to business and everyday applications.