AI-detection software looks for low levels of “burstiness” and “perplexity”. I.e. LLMs use sentences of even length that are easily grasped by the reader due to the use of predictable language patterns. This gives me a renewed appreciation of modernist literature & impenetrable academic texts! 😅