I did not read the paper, but what does more accurate mean in this case ? Likelihood of some unseen data ? Seems pretty hard to define or measure to me, if the goal is shear amusement.