whoahhhh, first time I've seen a model catch itself making a mistake outside of the reasoning mode