In deep learning in general, in GPT: some sensitivity to general exponent.