I agree with the notion, except that the models are indeed different

Some day maybe they will converge into approximately the same thing but then training will stop making economic sense (why spend millions to have ~the same thing?)