Large Language Models: A New Moore's Law?