What is a minimum number of tokens should be placed for a correct detection?
While testing your model I discovered that any small piece of text is detected as AI generated. What is a minimum number of tokens should be placed as text_example to see the correct results of the detector?
Thank you.
P.s.: Almost for any text (even created by me {with mistakes}), I see ~ 73% of AI generated content. Almost every piece of text extracted from a new articles on apnews.com gives me the same result. In rare cases I see that the detector is doubt about the one who made the content. And it's strictly rare when detector shows me "human generated".
My conclusion that is mostly produces false detects, unfortunately.
Hi alexcardo, most AI detector models that I know of prioritize recall over precision. Perhaps thats the reason you see a lot of false positives.