If you mean putting the model weights into gates directly, it’d be useless because users would get bored of the model as soon as they figured out what its style looked like. Also, large models can memorize their training data so eventually you’ll get it to output something copyrighted.
