Researchers at Apple have released a paper highlighting the limitations of AI models in mathematical reasoning. Their study reveals that even minor alterations in math problems can lead to incorrect conclusions from large language models (LLMs). For instance, adding irrelevant information about the size of kiwis confused state-of-the-art models, demonstrating their lack of genuine understanding. The researchers argue that LLMs replicate observed patterns rather than engage in true reasoning. This raises important questions about the capabilities and reliability of AI as it becomes more integrated into daily tasks.