[blah2244]

I've heard of a startup that claims to be able to achieve a near-0% false positive rate: https://www.pangram.com/our-model/how-it-works

They appear to basically RLHF a model on a bunch of examples of human/AI output on the same prompt. Not sure how well it works, but I'm guessing Mozilla is doing something similar here.

[saretup]

Anyone can get a 0% false positive by always inferring negative. What you wanna look at is the precision-recall curve.

https://en.m.wikipedia.org/wiki/Precision_and_recall