Because LLMs are language generation machines based on statistics - they do not analyse the underlying data, let alone understand it. They are not AI.