Recent research highlights the limitations of generative AI models, particularly large language models (LLMs). Despite their impressive performance in various tasks, these models lack a coherent understanding of the world. Researchers from MIT's Laboratory for Information and Decision Systems (LIDS) and other institutions have shown that even the best-performing LLMs do not form a true model of the world and its rules. This deficiency can lead to unexpected failures when the task or environment changes slightly. The findings, reported by MIT news and IntEngineering, suggest that while generative AI has advanced significantly, it still has a long way to go before it can be fully trusted. Additionally, the research raises questions about what aspects of generative AI are truly useful.