I don’t think this would work consistently, since creative work and inspiration take time and the detection errors might depend on the length of the work. I also think that making some groups of people jump through hoops before they can clear their name, because a generative algorithm said so, is only perpetuating the same problem.

Better would be to judge individual cases based on edit history + timestamps, and judging a tool based on multiple experiments like this. If it fails a few cases, then it should not be used.

Back when such “detectors” first popped up, our university sent out a guide to profs saying that they are unreliable and not to be trusted. We need similar education for more institutions since this isn’t common knowledge yet.