Heads up: this is technically not NSFW, but it will probably not look great if someone’s looking over your shoulder.
Not all that long ago, society was extremely anti-nerd. We bullied them, harassed them – force them into hiding in the various basements and Games Workshop stores in which they found refuge.
But we live in a more enlightened time now: we nurture these nerds, we encourage them. We give them all of the money we have so that they can make us movies about superheroes and apps that let you send videos of your awful dick to strangers.
They use their incredible intelligences and all the spare time they have because they don’t socialise to build incredible, bizarre, useless, beautiful tech innovations for us all to treasure, like this wonderful algorithm that can find genitalia in any landscape photo.
This next bit is a tad complicated so if you don’t really understand it, don’t worry, I don’t really either.
Developer Gabriel Goh got started when he got his hands on Yahoo‘s open source project open_nsfw, which is a neural network trained to identify “not safe for work” content. Automatically detecting NSFW content is a really hard problem to solve because it’s hard to tell what’s sexualised nudity and what isn’t, and it’s also suuuper expensive to hire people to do it for you.
Because it would basically be impossible to hand code what all the rules are for what constitutes rudity (new word I just made up for “rude nudity”), computer scientists (or “nerds”) fed a neural network thousands and thousands of rude images and let it make up its own rules about what a rude picture looks like.
Now that that’s been horribly oversimplified, let me explain what Goh did. While open_sfw is trained to look for NSFW content, Goh made a program based on Google‘s DeepDream that essentially generates images specifically to set off open_sfw’s NSFW detector. In offensively simplified terms: it’s a very complicated way of drawing dicks.
Combined with another program that does similar with harmless unsexy landscapes, Goh’s program takes the unsexy algorithm-generated picture and twists it into the other much more nightmarish algorithm’s idea of what vaginas and penises look like – and the results are fucking horrifying.
As an example, here’s what the algorithm reckons a beach looks like, compared to what the horny version reckons a beach looks like:
What looks mostly like a dream version of a concert gets turned into – let’s not beat around the bush here – a bunch of giant cocks:
While concerts seem decidedly phallic, for some reason it reckons deserts are yonic as hell:
Keep in mind, while this looks filthy as shit, these are not actually real pictures of anyone’s genitals, just a computer’s crazy hallucination of what genitals are.
The future is amazing.
Source and photo: Github / Gabriel Goh.
