@Amy in fairness, it’s trained on the output of humans who have also never managed to get it right.