fanfuchs:

chirotusinfinitum:

wolvensnothere:

monstermonstre:

wetwareproblem:

janothar:

elevengrackles:

madamehearthwitch:

bnq:

janothar:

sorensfw:

What is almost certainly happening with Tumblr’s image classifier is that they’ve deployed the bot entirely untrained, and are relying on user complaints to train it what NSFW content on tumblr looks like.

Making people participate in their own censorship in this way and punishing them with the threat of being silenced even if in compliance with the rules should they not participate, is, IMHO, abuse of their userbase.

Source: I’ve worked for three different top-five social networks, and at all of them this type of work would have been done by making the userbase do it as much as possible.

Please signal boost. I don’t care if you reblog or repost, just get the word out so that people know what they’re doing as they interect with this thing.

My twitter is @soren_tycho (nsfw, sfw account coming soon) and my github username is sorentycho.

Just chiming in as a machine learning expert, this is almost certainly the case (and knowing tumblr, they weren’t even smart enough to throw in pre-trained weights)

I don’t remember agreeing to beta test 🤔

A well known AI blogger I follow also supports this theory.

Okay so what I am hearing here is that if we really want to be assholes and fuck with tumblr then we should ask for review even on flagged posts that are legitimately NSFW?

Fuck with their robot training/ using us as laboratory rats by being the worst possible laboratory rats?

@sorensfw @janothar

The review process will catch it and clean the data, but it WOULD put more strain on whatever humans they have there. Though those humans won’t be the people responsible, they’ll be minimum wage people, so…

Catch: there is mounting evidence that there aren’t any human reviewers – most notably, I’ve seen a post about a basket of mushrooms being rejected on appeal.

Confusing matters even more, I’ve had two copies of visually identical images
handled differently – an unedited version approved, and a version with text steganographically embedded in it rejected.

@wetwareproblem I agree: I appealed an actual nsfw post as protest of their new policies and aggressive untrained bots and got a response apologising for the “false flagging” and saying it respected community guidelines

the response also came faster than any legit sfw posts (usually queer content) I’ve appealed in the past, before the nsfw ban

a human didn’t review this, or if they did, they are overworked and understaffed

So. Hey. If this is what’s happening? It would quite possibly be the stupidest fucking thing tumblr has yet ever done.

I fully support a massive campaign of intentionally flagging all images and disputing the flagging of all images including actually NSFW images. Let’s see of we can break this thing.

Soooo we could train the flag bit and the appeal bot to leave consensual porn alone but get rid of nazis, maps, spam bots and exploitative porn?
Problem is: all of those latter probably already clued in to this, so this basically becomes an arms race…