Today’s AI-generated prose is riddled with flaws. Chatbots produce meaningless metaphors, endless “it’s not this, but that” constructions, and a cloyingly sycophantic tone—and, of course, they overuse my beloved em dash. (Only starting with GPT-5.1, released in November, could ChatGPT reliably follow instructions to avoid the beleaguered punctuation mark.) I wanted to understand why this is—why large language models, which, after all, have memorized centuries of great literature, can demonstrate incredible emergent abilities yet totally fail to produce a single essay that I’d want to read.