Its the problem with fully proprietary AI like this: You cannot prove that this question and this answer wasnt in the training set, so you cannot argue for its ability to infer or reason.