Comment by amelius

In that case, shouldn't the training data be open source? Seems like somebody got an unfair advantage here.