autismserenity:

allthoseotherworlds:

telegantmess:

shadowpeoplearejerks:

kirby-ebooks:

shadowpeoplearejerks:

1999msi:

lord-of-the-discourse:

First and foremost: IT HAS BECOME VERY CLEAR TO ME THAT LGBT CONTENT IS NOT INTENTIONALLY BEING FILTERED OUT

I checked several LGBT related tags and the tags were still full of content with the filter on. I checked several porn tags, and they were totally empty. It’s legitimately meant to only filter adult content, anything else being marked as sensitive content is a bug and not done by design.

You can also mark something as sensitive when you post it

The same thing happened with Youtube. Filtering content is NOT EASY, there were bound to be problems like this. 

one of the problems of “big data” and anything involving trying to extract meaning from language through computers is that the state of the art technology (so-called AI, deep learning, neural nets) tend to recreate bias if there is any bias in the data they are trained on. What happened with youTube was pretty clearly that they were going by “what are the attributes of videos that get reported a lot?” and the neural net noticed that videos with lgbt content tend to get reported a lot. so the neural net assumes, because it is a dumb robot with secret methods, that lgbt content is offensive! it’s almost definitely the same thing going on with tumblr, but probably with worse algorithms, because tumblr. 

it’s probably not “malicious” in that probably an engineer didn’t sit down and write out “gay = bad” in a line of code, but it is negligent in that they didn’t CHECK if their algorithm replicated existing bias. this is a continual fucking problem in the field and it’s irresponsible and bad engineering to fail to PLAN FOR IT. 

That’s true. Engineers in machine learning should be well aware of this issue at this point and it’s pretty embarrassing that they let things out into the wild without fixing this. 

“Known issue with past iterations of this kind of car, cars’ brake pads randomly burst into flames”
“Eh, I’m sure we did it right. Roll it out!”

“it’s probably not “malicious” in that probably an engineer didn’t sit down and write out “gay = bad” in a line of code, but it is negligent in that they didn’t CHECK if their algorithm replicated existing bias. this is a continual fucking problem in the field and it’s irresponsible and bad engineering to fail to PLAN FOR IT.”

just repeating this part for emphasis

I feel like this is what what the testing phase of development is supposed to notice

Oh, it is.

I’m gonna go ahead and guess that they don’t have a QA department, AND don’t require that devs do it.