It's well known that LLMs don't reason. That's not what they are for. It's a throw away comment to say that a product can't do what it explicitly is unable to do. Reasoning will require different architectures. Even with that LLMs are incredibly useful.