Hi @Copyleaks ,
I appreciate you engaging in this discussion. I read through the limitations in that web page, and it was interesting. What I want to know, which has been most difficult to find, is not just the false positive rate overall, but the differential false positive rate for specific populations that have been shown to have higher rates, such as English Language Learners and some neurodivergent populations who may have more idiosyncratic writing styles. Has your company done any analysis on that?
It's not sufficient, for example, to know that there is an overall .2% rate, if there are populations that have significantly higher rates. If you're already a marginalized student, and you continually get incorrectly flagged (and punished) for AI use, it can be the "last straw".
NOTE: The captcha is getting ridiculous. It went through about 25 challenges and then failed verification. The accessiblity on this is ridiculous. I have both visual and auditory disabilities, and making me go through this many hoops is pretty much discouraging me from participation in this community.