“According to OpenAI, the classifier incorrectly labels human-written text as AI-written 9% of the time.”
OpenAI’s detector catches AI text about 26% of the time and falsely flags humans 9% of the time. It needs at least 1,000 characters and does worse on text written by children, the exact people schools will run it on. Changing a few words beats it. OpenAI sells the poison and now hands out a placebo antidote.