If its detecting non-native English as GPT produced then wouldnt that mean GPT is modeled on non-native English to produce similar outputs? If it is NOT detecting native English as GPT produced then wouldn't that mean it's showing up different than the training data outputs?