4 ms·
(haven't read the paper) Doesn't it boil down to a halting-problem kind of argument? Assume you have a model that can detect AI-generated text. Then you can b
by puttycat 4y ago
(haven't read the paper)
Doesn't it boil down to a halting-problem kind of argument?
Assume you have a model that can detect AI-generated text. Then you can build a model that emulates the detection model, finds the nearest non-detected text, and outputs it.
- sebzim4500 4y agoTrue but that could be computationally impractical, depending on how hard the detection process is.
- Atlas22 4y agoCouldnt the same be done with the generation model? I.e, The detection model can emulate the generation model to inform detection.