7 ms·
As I understand it, the current watermarking methods rely on a secret key, making the detection schemes a black box to anyone not in possession of the key. Thi
by dare944 1mo ago
As I understand it, the current watermarking methods rely on a secret key, making the detection schemes a black box to anyone not in possession of the key. This means organizations like Anthropic are free to make any claim about authorship they want, true or not, and no one can call them on it.
- demibabs 1mo agoPart of the legislation requires them to make a public AI text detector (ala GPTZero I assume). Wouldn’t having that be enough to eventually reverse engineer the key?
- dare944 1mo agoNot if they designed the algorithm right.
- inigyou 1mo agoProbably not to get the key, but you could certainly use it adversarially to remove the watermark. Removal may come down to changing every third token to a different one.
- NewsaHackO 1mo agoYea, they are going to have to monitor how often similar writing pieces are being submitted, or limit the access to the checker somehow.
- inigyou 1mo agoEurope trusts its institutions way too much. It may well be that only the police can access the checker.