Because they do not have any intelligence and just predict next tokens, which is a crappy method for determining answers. Because they are not deterministic and

Because they do not have any intelligence and just predict next tokens, which is a crappy method for determining answers. Because they are not deterministic and can (and will) give different people different answers, and even the same person different answers at different times. Because, in short, LLMs suck as tools and shouldn't be relied upon for anything important.