This is like saying a dictionary knows more words than a human author. The premise is highly conditional. LLMs are tools that answer questions in human language. They aren't physicians. So the only way they can outperform a human doctor at this point is "on paper" only.