News for hackers. Stuff that matters.
Re:Tumblr has lost 30 percent of web traffic since December by dgzl on Thursday March 14, 2019 @07:37PM (#19395010)
I imagine it would take careful and meticulous observation to differentiate between legal porn and almost-legal porn. Dissolving the platform of porn entirely would streamline this process and potentially be automated. Banning regular porn has a greater than zero bearing on moderating child porn, and you're being disingenuous in pretending otherwise.
Re:Tumblr has lost 30 percent of web traffic since December by chimeracoder on Thursday March 14, 2019 @10:10PM (#19395843)
> I imagine it would take careful and meticulous observation to differentiate between legal porn and almost-legal porn. Dissolving the platform of porn entirely would streamline this process and potentially be automated. Banning regular porn has a greater than zero bearing on moderating child porn, and you're being disingenuous in pretending otherwise.

Why do people keep repeating this?

No, it's much more difficult to ban all pornography than it is to ban only child porn. One requires analyzing the content of photos and the other doesn't.

Re:Tumblr has lost 30 percent of web traffic since December by dgzl on Thursday March 14, 2019 @10:38PM (#19395962)
> No, it's much more difficult to ban all pornography than it is to ban only child porn. One requires analyzing the content of photos and the other doesn't.

Since it's easier to spot pornography in general than differentiate between very disturbing types of pornography, the work load is lessened and overall less stressful. Not to mention the amount of work ai could automate. That's your point, right?

Re:Tumblr has lost 30 percent of web traffic since December by chimeracoder on Thursday March 14, 2019 @10:54PM (#19396036)
> Since it's easier to spot pornography in general than differentiate between very disturbing types of pornography, the work load is lessened and overall less stressful. Not to mention the amount of work ai could automate. That's your point, right?

No, detecting child pornography by matching photos to known datasets and then extrapolating rings dedicated to sharing child porn from the social graph is already a solved problem. NCMEC already provides the tools to do this. You don't need to do any image analysis.

There are no effective tools for detecting porn more generally. That's a much harder problem.

Re:Tumblr has lost 30 percent of web traffic since December by michaelt on Friday March 15, 2019 @07:02AM (#19398497)
According to [1] a Tumblr spokesperson said "we work collaboratively with [...] partners like NCMEC to actively monitor content uploaded to the platform. Every image uploaded to Tumblr is scanned against an industry database of known child sexual abuse material, and images that are detected never reach the platform."

Evidently that wasn't sufficient to stop Apple removing them from the App Store, though.

[1] https://www.theinquirer.net/inquirer/news/3066654/tumblr-pul...

Re:Tumblr has lost 30 percent of web traffic since December by monadgonad on Friday March 15, 2019 @07:10AM (#19398545)
Most of the child porn probably wasn't known stuff shared by rings, but teenage users uploading their own photos.
Re:Tumblr has lost 30 percent of web traffic since December by chimeracoder on Friday March 15, 2019 @07:59AM (#19398826)
> Most of the child porn probably wasn't known stuff shared by rings, but teenage users uploading their own photos.

Except that's not what the issue was.

Re:Tumblr has lost 30 percent of web traffic since December by darkpuma on Friday March 15, 2019 @03:26AM (#19397386)
You're blatantly wrong. Training a classifier to determine if a previously unseen image contains nudity is easier than training a classifier to determine if a previously unseen image contains a nude 18 year old or a nude 17 year old. The later task is virtually impossible, which should make it obvious to you that tech like PhotoDNA is not classifying previously unseen photographs.
Re:Tumblr has lost 30 percent of web traffic since December by chimeracoder on Friday March 15, 2019 @06:08AM (#19398233)
> You're blatantly wrong. Training a classifier to determine if a previously unseen image contains nudity is easier than training a classifier to determine if a previously unseen image contains a nude 18 year old or a nude 17 year old.

Why is it so hard to believe that there exists a solution to this that's much easier than training an image classifier?

As I've explained elsewhere, multiple times, this is a solved problem. There's no image analysis required.

Re:Tumblr has lost 30 percent of web traffic since December by darkpuma on Friday March 15, 2019 @10:50AM (#19400189)
> previously unseen

You missed that part. That requires classification, either by a human or by an algorithm. There is no way around it.

Re:Tumblr has lost 30 percent of web traffic since December by chimeracoder on Friday March 15, 2019 @12:05PM (#19400919)
> You missed that part. That requires classification, either by a human or by an algorithm. There is no way around it.

I did not miss anything. "Previously unseen" only matters if your primary means of identifying people who share child porn is based on analyzing the content of images.

I'm very aware of how these systems work, not just at Tumblr but in the industry at large at other, larger companies. I'm not just speculating here; this is literally how NCMEC works and what they expect of companies that host user-generated content.

Re:Tumblr has lost 30 percent of web traffic since December by darkpuma on Friday March 15, 2019 @12:15PM (#19401019)
For years tumblr neglected to address the issue of teenagers that haven't been trafficked or otherwise gone missing, posting pictures of themselves that did not exist in any database. That is the issue they addressed by banning all pornography, removing any ambiguity over whether any particular never-seen-before image was permitted or banned. NCMEC is out of scope here. They do not have magical powers that allow them to distinguish between an 18 year old or a 17 year old uploading pictures of themselves.

You may not like it, I can tell you're very upset at tumblr banning pornography, but this is the reality. Throw around as many of your qualifications as you like, it changes nothing.

Re:Tumblr has lost 30 percent of web traffic since December by chimeracoder on Friday March 15, 2019 @12:34PM (#19401277)
> NCMEC is out of scope here. They do not have magical powers that allow them to distinguish between an 18 year old or a 17 year old uploading pictures of themselves.

NCMEC doesn't need those "magical powers" because NCMEC does not care about the latter. That's not their mandate. Hacker News loves to talk about age-of-consent and child pornography in the context of 17 year olds, but that's simply not relevant here. That's not the issue that Tumblr was facing pressure from.

> Throw around as many of your qualifications as you like, it changes nothing.

It sounds like you're so convinced in your own beliefs that even direct, firsthand knowledge of the inside situation won't change your mind, in which case you're right - there's no point in continuing this conversation.

You may not like that I know what actually went on here, and I can tell you're very upset that the actual facts about how NCMEC works and why Tumblr internally chose to make their decision don't support your narrative, but that's the reality. Whether you choose to believe it or not, it changes nothing.

Re:Tumblr has lost 30 percent of web traffic since December by darkpuma on Friday March 15, 2019 @01:53PM (#19402295)
> "because NCMEC does not care about the latter."

And nor do you it seems, but tumblr now does. They've been forced to. You may not like it, but that's the reality of how tumblr is now operating. Tumblr can no longer turn a blind eye to any form of child pornography, and is no longer turning a blind eye to any form of child pornography.

Tumblr didn't ban all pornography for shits and giggles. They did it because given the current legal and social landscape it was their only economically viable option.