Once, I encountered the funny story of an AI image descriptor with a sheep obsession. It had been trained on pictures of fields of sheep. Therefore, it tagged anything in a field as ‘sheep’, including an empty field, because they work on statistical probability. Therefore, it thinks “ah, a field! there’s probably a sheep here.” (It’s a bit more complicated but basically that.) It also couldn’t recognise sheep in places that weren’t fields, such as petrol stations or barns. [cont]

solivar:

garbage-empress:

crazy-pages:

crazy-pages:

retroactivebakeries:

oneiroteuthis:

lauralot89:

Now, the alarming aspect of this story is that the very same technology is probably what tumblr is using to identify porn. Now, if it can’t tell that an empty field is not, in fact, full of sheep, what hope do we have that it can’t tell an empty room isn’t full of writing human forms engaged in passionate coitus?

this really does sound like an episode of black mirror

But wait, it’s even weirder than that!

This is gonna produce some absolutely baffling pornography.

…. oh my fucking god they actually are using open source software. They’re using a fucking one-layer unidirectional bicategory tag-trained neural network. This will never work. Literally, it will never work. There’s just not enough algorithmic complexity to do what they’re asking of it. I bet you I could prove on a mathematical level that this joke of a neural net fundamentally lacks the abstraction necessary to do its job.

This will never get better. Their algorithm will never stop fucking up, it will never actually flag porn reliably and it will always require a massive quantity of human hours to deal with the deluge of mistagged pictures. This isn’t just a case of an insufficiently trained algorithm, it’s just … this is the most basic neural network you can make. It probably hasa a lot of neurons and has loads of training data but like … you can’t just brute force this kind of stuff. One layer of neurons is just Not Enough.

Also, just to make this clear, Tumblr lied. I mean, we already know this, but I mean they liiiieeeeed. All that stuff they promised about what would or would not be censored? That cannot be delivered on with a system this simple. Nude classical sculptures, political protests, male-presenting nipples (really Tumblr?), nude art outside the context of sex, all that? You cannot train a bicategory one-layer neural network to exclude those things. It cannot be done. Tumblr never intended for those things to actually be permitted, they were just lying. Because the system they have cannot actually do what they said it would and never will be able to.

Also, this kind of system is super vulnerable to counter-neural strategies. I bet you before the end of the month someone hooks up their own open source one layer bicategory neural network which puts an imperceptible (to humans) layer of patterned static over arbitrary images, and trains it by having it bot-post static-ed images to Tumblr and reinforcing based on whether the images are labeled nsfw or sfw. Seriously, within a month someone will have an input-output machine which can turn any image ‘sfw’ in Tumblr’s eyes.

This is genuinely pathetic. Like, I have real pity for whoever implemented this, because it’s clear Tumblr doesn’t actually have any engineers with any expertise with machine learning left at all and they foisted the job off on some poor bastard who has no idea what they’re doing and is going to get all kinds of flak for their (perfectly reasonable and predetermined) failure from management.

As has been pointed out before, there are no humans behind this at all. The review process just reruns either the same algorithm or another algorithm, but people have posted screen shots showing obviously SFW pictures that were still deemed NSFW on review, despite the fact that any human, no matter how overworked / tired would have seen that these pictures were not porn.

Meanwhile, yesterday, I received a notification indicating that my blog was the origin of actual live-action porn gifs – which were not actually on my blog but the origination tag nonetheless redirected back to my most-heavily-reblogged extremely SFW/containing no images at all protest post.

Censored on tumblr

lovesicksick:

lovesicksick:

ask-a-yandere:

lovesicksick:

What all ask boxes look like for me (submission boxes look exactly the same):

Left: what happens when I try to open dms (New Message does nothing); I can’t even read them. Right: when I try to go to someone’s dms from their blog.

Neither new things I post nor old things I post show up in the tags they’re in, even if they were before.

I can type things into replies, but it won’t actually post them.

Since it doesn’t seem this one has been encountered by many people yet, maybe reblog this so people know about it. I’m probably getting deleted though

Tumblr is shadowbanning people and they aren’t even trying to be subtle about it.

Please spread the word.

I’m glad this is spreading but simultaneously it makes it sound like the shadowban is much less severe than it is because I didn’t realize, and this shadowban is so severe I can’t even add on / tell anyone that.

All my likes, reblogs, and replies are hidden from the notes of EVERY post.

When I like or rb things it doesn’t show notes in someone’s activity / dash, except long-time followers (not sure about the exact cutoff/conditions).

If I reblogged and added something to this from a non-follower they wouldn’t even know.
Check and see: my rb’s don’t even show up in the notes.
So people couldn’t even find out I added something on by checking the notes either:

Not to mention: I’m hidden from all searches, can’t be @’d, past @’s of me are hidden from my notes, I don’t get notified when people reply to a post I replied to, my @’s don’t give notes, and every search on my blog is blanked. I can’t even ask/submit to myself even with anons, and lots more.

Oh, yeah, and: MY BLOG ISN’T EVEN MARKED NSFW. Every flagged post I’ve found was restored. But every flagged post you contest brings attention to you that risks you just getting shadowbanned randomly.

THIS is the version that should get spread around…

apricops:

apricops:

It’s quite likely no coincidence that that most ‘mismanaged’ and least profitable social media site is also the one that turned out to be most amenable to the formation of actual communities

To clarify, Tumblr is indeed horribly mismanaged, but notably, it’s mismanaged both in ways that harm us (e.g. doing little about pornbots, nazis, etc.) and ways that have greatly benefited us – not asking for real names, hiding our follower counts, a chronologically-sorted dashboard, etc. are big draws, but in the eyes of other social media monarchs, they look like unforgivable mistakes. If I don’t have to give my real name, that’s that much less information to sell to advertisers. If posts are listed chronologically, Tumblr can’t shove the posts of ‘influencers’ in front of me willy-nilly. Tumblr was a ‘success’ because it was too poorly managed to sufficiently atomize us, and so we actually had conversations and communities instead of being the best products for advertisers.