Paul Krugman: As far as I can tell, large language models — which we are, misleadingly, calling artificial intelligence — are still, essentially, a souped-up version of autocorrect.