Despite its impressive output, generative AI doesn’t have a coherent understanding of the world: « Researchers show that even the best-performing large language models don’t form a true model of the world and its rules, and can thus fail unexpectedly on similar tasks. »