A new study published in Nature reveals that AI systems are developing an unexpected ability to generalize knowledge across different domains, a trait previously thought to be uniquely human. Researchers from several leading institutions trained models on specific tasks, such as image recognition, and found the systems could apply learned patterns to unrelated areas like language translation without additional training. This emergent cross-modal generalization suggests AI may be developing more fundamental learning principles than its creators intended. While promising for creating more versatile AI, the phenomenon raises questions about predictability and control as systems exhibit capabilities beyond their initial programming parameters. The full findings are detailed in the article: https://technologyreview.com/2024/05/ai-generalization-study.