AI loves negative parallelism

Anthropic recently announced it's adding watermarks to all text outputted by its chatbot Claude. But often, you don't need fancy detection software to know; you just need to look for language that gets repeated over and over by AI chatbots, like “it's not X, …