Are large language models a good model of *human* language?

garymarcus.substack.com