I picked two random numbers between one and one million. The chances of it having seen that specific problem in its training set seem very low.