AI model demonstrates flawless ability to be wrong about being right