Luca Ambrogioni, Assistant Professor of Machine Learning at Radboud University, challenges claims that AI detectors can confirm fully automated writing in the recent Granta fiction scandal

I don't think that this is true. These detectors can only reliably isolate the final level of writing They cannot distinguished self generated writing from highly iteratively prompted and edited writing A 100% flag means that an LLM was highly used in the writing process, not that the writing was written without a human in the loop