News for hackers. Stuff that matters.
Re:Tumblr has lost 30 percent of web traffic since December by TorKlingberg on Thursday March 14, 2019 @07:05PM (#19394789)
Tumblr didn't do it because they wanted to. They were forced by Apple removing their app from the App Store.
Re:Tumblr has lost 30 percent of web traffic since December by vilhelm_s on Thursday March 14, 2019 @08:41PM (#19395370)
Apparently they had already decided to do it before the App Store thing happened.

https://www.vox.com/the-goods/2018/12/4/18126112/tumblr-porn...

> But a former staff engineer, who recently left Tumblr and asked to remain anonymous for professional reasons, tells Vox that the NSFW ban was “in the works for about six months as an official project,” adding that it was given additional resources and named “Project X” in September, shortly before it was announced to the rest of the company at an all-hands meeting. “[The NSFW ban] was going to happen anyway,” the former engineer told me. “Verizon pushed it out the door after the child pornography thing and made the deadline sooner,” but the real problem was always that Verizon couldn’t sell ads next to porn.

Re:Tumblr has lost 30 percent of web traffic since December by iknowstuff on Thursday March 14, 2019 @08:57PM (#19395467)
Why do companies refuse to have ads on a well known website on which some posts happen to be porn? Is it an american thing?
Re:Tumblr has lost 30 percent of web traffic since December by jdavis703 on Thursday March 14, 2019 @09:27PM (#19395653)
It’s not an American thing, most advertisers also refuse to advertise next to content that references extreme violence (think ISIS beheadings) and gambling content. I worked on the ads team at a company, and we constantly had to tweak our filters to ensure ads didn’t get served next to “adult” content.
Re:Tumblr has lost 30 percent of web traffic since December by subpixel on Thursday March 14, 2019 @09:33PM (#19395678)
Meanwhile I have to watch a pre-roll ad to see reportage about a terrorist attack.
Re:Tumblr has lost 30 percent of web traffic since December by pjc50 on Friday March 15, 2019 @06:12AM (#19398245)
> most advertisers also refuse to advertise next to content that references extreme violence

Apparently the Daily Mail Online is running this morning's NZ murder video on its front page. Right next to the ads.

Re:Tumblr has lost 30 percent of web traffic since December by chimeracoder on Thursday March 14, 2019 @10:12PM (#19395854)
> But a former staff engineer, who recently left Tumblr and asked to remain anonymous for professional reasons, tells Vox that the NSFW ban was “in the works for about six months as an official project,

Ah, so shortly after SESTA was signed into law.

Re:Tumblr has lost 30 percent of web traffic since December by eridius on Thursday March 14, 2019 @11:00PM (#19396062)
Tumblr's porn ban has literally nothing at all to do with the App Store. The porn ban is because of SESTA/FOSTA.
Re:Tumblr has lost 30 percent of web traffic since December by philpem on Friday March 15, 2019 @06:15AM (#19398260)
Sounds like it was planned much earlier:

https://news.ycombinator.com/item?id=19395370

Re:Tumblr has lost 30 percent of web traffic since December by favorited on Thursday March 14, 2019 @07:56PM (#19395123)
That was (according to reports) because of the prevalence of child pornography on their platform, which they apparently didn't have the tools or staff to police. A blanket ban on pornographic content was cheaper and/or easier.
Re:Tumblr has lost 30 percent of web traffic since December by deogeo on Thursday March 14, 2019 @08:14PM (#19395219)
Presumably child porn was already banned on Tumblr, so how does banning all other porn as well help?
Re:Tumblr has lost 30 percent of web traffic since December by retsibsi on Thursday March 14, 2019 @08:19PM (#19395242)
I believe they're using some automated filtering (see e.g. https://www.wired.com/story/tumblr-porn-ai-adult-content/). I assume it's easier to train an AI to recognise pornography generally, rather than a specific kind -- and now that I think about it, the process of training an AI to recognise child porn sounds extremely unpleasant and legally dubious.
Re:Tumblr has lost 30 percent of web traffic since December by chimeracoder on Thursday March 14, 2019 @10:06PM (#19395819)
> I believe they're using some automated filtering (see e.g. https://www.wired.com/story/tumblr-porn-ai-adult-content/). I assume it's easier to train an AI to recognise pornography generally, rather than a specific kind -- and now that I think about it, the process of training an AI to recognise child porn sounds extremely unpleasant and legally dubious.

Quite the opposite - it's much easier to allow pornography and ban only child pornography than it is to create an automated system to detect pornography.

PhotoDNA makes it easy to find matches, which can then be used to uncover the networks of people posting child pornography, which are then added back to the database. It doesn't require any computer vision at all.

By contrast, banning all pornography does require computer vision of some sort, and that's much more difficult, as evidenced by how terrible the new Tumblr NSFW content detector is.

Re:Tumblr has lost 30 percent of web traffic since December by retsibsi on Thursday March 14, 2019 @10:40PM (#19395972)
Maybe I'm misunderstanding, but PhotoDNA seems to be a tool for identifying reposts and edits of already known illegal images -- those that have previously been found and had their hashes entered into the database. So if Tumblr had a problem with original content (which could include underage users posting explicit pictures of themselves, as well as the worse things we think of when we hear the phrase 'child porn'), I don't think that would help.

edit: I found this quote from an article[1] published last November: "In its updated statement, Tumblr said that while every image uploaded to the platform is “scanned against an industry database of child sexual abuse material” to filter out explicit images, a “routine audit” discovered content that was absent from the database, allowing it to slip through the filter."

So it looks like they might have already been using PhotoDNA.

[1] https://www.theverge.com/2018/11/20/18104366/tumblr-ios-app-...

Re:Tumblr has lost 30 percent of web traffic since December by chimeracoder on Thursday March 14, 2019 @11:07PM (#19396094)
> So it looks like they might have already been using PhotoDNA.

Of course they were working with NCMEC (and therefore using PhotoDNA, which is owned by NCMEC); it would have been legal suicide for them not to. They were doing this for years, before they were even acquired by Yahoo.

> Maybe I'm misunderstanding, but PhotoDNA seems to be a tool for identifying reposts and edits of already known illegal images -- those that have previously been found and had their hashes entered into the database. So if Tumblr had a problem with original content (which could include underage users posting explicit pictures of themselves, as well as the worse things we think of when we hear the phrase 'child porn'), I don't think that would help.

How many people do you think have Tumblr accounts where they post only new, never-before-seen pornography of underage children, which has never been posted on any other blog before, and never once post a single photo that has been previously identified as child pornography[0]?

Of those, how many do you think are able to ensure that none of their followers ever repost/reblog that photo on any other blog which also happens to contain at least one other photo that's been previously identified as child pornography?

Of those, how many do you think are able to ensure that none of their followers have been previously identified as highly-connected nodes in the underground networks dedicated to sharing child pornography, and therefore pretty much exclusively post child pornography or follow people who they believe are likely to post child pornography?

Of those, how many do you think are able to ensure that nobody who ever sees one of those photos ever decides to download it and upload it as an attachement to an unsent email in their Gmail drafts folder, or put it in a private Dropbox folder (shared with nobody), or sends it on any of the many "cloud" services which also actively monitor for child pornography and do the same sorts of graph analysis to identify people who are using their services to store or share child pornography?

Again, once you understand how these underground networks work, and once you realize that this is mostly a problem of social graph analysis and not image recognition/classification, you realize that it's very easy to solve.

[0] Remember, not just new to Tumblr, but new to everyone who is working with NCMEC/ICMEC (which means every large and not-so-large company in the entire world that hosts user-provided content).

Re:Tumblr has lost 30 percent of web traffic since December by retsibsi on Thursday March 14, 2019 @11:21PM (#19396160)
You are obviously better informed than me, but if the problem is so easy to solve, and they were already using the relevant tool, why was it still a problem? Are you arguing that they failed through incompetence, or that they were lying/exaggerating as an excuse to ban all pornography for other reasons?
Re:Tumblr has lost 30 percent of web traffic since December by chimeracoder on Thursday March 14, 2019 @11:31PM (#19396213)
> Are you arguing that they failed through incompetence, or that they were lying/exaggerating as an excuse to ban all pornography for other reasons?

Yes, SESTA was the real reason that they banned pornography. SESTA imposed way too much liability on them to be able to support it.

It had nothing to do with child porn. They are still liable for child porn. They still have to work with NCMEC and go through the same process for identifying, reporting, and removing child porn. None of that has changed.

Re:Tumblr has lost 30 percent of web traffic since December by snvzz on Friday March 15, 2019 @02:50PM (#19402921)
So, if I understand correctly, this means that by banning all pornography in the name of preventing child porn, they'll actually be harming efforts on this uncovering of child pornography networks effort.

This is unfortunate.

Re:Tumblr has lost 30 percent of web traffic since December by chimeracoder on Friday March 15, 2019 @03:05PM (#19403069)
> So, if I understand correctly, this means that by banning all pornography in the name of preventing child porn, they'll actually be harming efforts on this uncovering of child pornography networks effort. This is unfortunate.

Unfortunate, and also wholly predictable. This was exactly why advocacy groups of former victims of sex trafficking opposed SESTA in the first place, as did even the DOJ: they both recognized that it would ultimately make it harder to identify and rescue victims of sex trafficking.

Re:Tumblr has lost 30 percent of web traffic since December by protomyth on Thursday March 14, 2019 @08:18PM (#19395238)
Flagging all skin is much easier than determining age of person showing skin.
Re:Tumblr has lost 30 percent of web traffic since December by Dylan16807 on Thursday March 14, 2019 @09:19PM (#19395598)
If they cared about making the filtering easy then I'm not seeing why they banned drawings too.
Re:Tumblr has lost 30 percent of web traffic since December by protomyth on Thursday March 14, 2019 @11:10PM (#19396102)
If you look at a lot of this IDing software, it grabs drawings too. Add some ok drawings in Photo in iOS and watch it add it to the faces list.
Re:Tumblr has lost 30 percent of web traffic since December by Dylan16807 on Thursday March 14, 2019 @11:46PM (#19396294)
It can but it makes the classification job significantly more complicated.
Re:Tumblr has lost 30 percent of web traffic since December by protomyth on Friday March 15, 2019 @12:23AM (#19396519)
Interesting, because it seems like it will pick up drawings for free because it doesn’t distinguish between photos and drawings (Photo on iOS for example will pick up line drawings of people). I actually thought it would a harder problem to exclude the drawings based on some of these facial id programs.
Re:Tumblr has lost 30 percent of web traffic since December by Dylan16807 on Friday March 15, 2019 @12:29AM (#19396556)
A lot of drawings don't look at all like photos, and they are banned too. That's definitely not for free.
Re:Tumblr has lost 30 percent of web traffic since December by dragonwriter on Thursday March 14, 2019 @08:18PM (#19395241)
Since people don't go through a visually obvious physical change at the moment of attaining legal adulthood, banning all porn leaves less cracks for child porn to get through.

I'm not saying it's necessarily the right policy, but it does.make a difference.