<i>"The models particularly struggle with open-ended diagnostic reasoning.”</i><p>The models struggle with reasoning ... period.<p>Any apparent reasoning ability is really a statistical simulation thereof drawing on similarity with training data.