3 ms·
Facebook has been using ML to detect “revenge porn” images since at least 2018. Likely they don’t know any more than you or intone else does why their ML class
by efitz 4y ago
Facebook has been using ML to detect “revenge porn” images since at least 2018.
Likely they don’t know any more than you or intone else does why their ML classifies your image that way. In fact, one of my pet peeves about ML is that most of the time no one can meaningfully explain why it made any particular decision (they can explain inputs and weights and the algorithms used and matrix math, etc, but can’t say “the model was 93% sure that part of the image was a penis”).
Anywhow if an actual human did look at your first appeal, you did not do yourself any favors calling them “dense”. People don’t like it when you call them stupid[1].
[1] https://www.theonion.com/people-dont-like-it-when-you-call-them-stupid-1819583684 https://www.theonion.com/people-dont-like-it-when-you-call-t...
- jrm4 4y agoYou're not wrong about the "dense" thing, but it sure highlights the need to keep examining and reexamining the relationship between services like Twitter and e.g. censorship, governance, and the First Amendment. Nobody's breaking the law here, nobody's being unreasonable, I think, but this isn't a great balance of power.