Better language models and their implications