Comment on Are there any privacy-friendly (including offline) AI detectors?
sukhmel@programming.dev 9 months ago
The problem with those detectors is that they are barely better than guessing.
You’re most likely capable of outperforming such a detector by spotting semantic issues. Like when I tried to find how to configure saddle stool and the article said that my outstretched legs should reach pedals.
That may also be due to poor and cheap manual writing but you’ll not lose a lot by blocking those, I guess.
Showroom7561@lemmy.ca 9 months ago
I’ve tested quite a few, and they seem extremely accurate. Even when fragments of text have been AI generated, they point those out.
I mean, some websites I’ve visited seem AI generated, and the verification sites simply confirm my suspicions.
Lemongrab@lemmy.one 9 months ago
Test them on known good articles, like those before 2019, and see the rate of false positives.
Showroom7561@lemmy.ca 9 months ago
0% AI detected.
It does rate on a scale, so it’s not always 100% or 0% (sometimes it is). But it has been very reliable in the tests I’ve done. Apparently, false positives are something like 0.2%.