Haha they have a lot of spam to cleanup then. A frequent discovery of mine is a new foothill of the mountains of spam text dumped onto GitHub. No idea what hal

Haha they have a lot of spam to cleanup then. A frequent discovery of mine is a new foothill of the mountains of spam text dumped onto GitHub. No idea what half of it is there for since it’s not even connected to a GitHub pages setup most of the time. Usually I’ll find them a few pages back for a code search where the spellchecker has helpfully ignored pluralisation and punctuation and I’m getting lots of generic results but for a long enough string of words to narrow down GitHub to the long tail of noise… Lots of txt and markdown and html and textish formats filled with keywords and phrases, some of which are surprisingly suggestive, arguably outright pornography… Oh and there’s the matter of all the code on GitHub related to adult websites and the pornography that they contain, heaps of web scraping scripts to harvest videos and pictures from “tube websites” … the flip side of the coin as well, the content moderation tools like perceptual hashing tools and ML models for adult content or nudity detection that need to include sample images as part of testing and training data… This feels like an area GitHub really shouldn’t be fucking with with respect to the whole “what counts as acceptable”, it’s one thing to prevent people using GitHub to host a free porn site via GitHub pages, it’s another to start policing the definition of acceptable discourse on the topic of adult content and machine learning models, particularly given that machine learning is a highly technical topic and the state of the art is an evolving field full of deeply technical people. I’m deeply unimpressed by the results from stable diffusion but I do see other people having success with it. For me even the most basic things seem to come out mutated and the same efforts to improve non pornography related human depictions from stable diffusion type models will also be applicable to porn … if the hands on my “18th century porcelain doll of a cowboy“ look like some kind of teratogenic mutant out of a bad hyper realistic cartoon, then the porn version can’t be much better and while rule 34 exists the majority of “pornography consumers” looking to ML model based content probably don’t want that either which leads to a sort of “dual use technology” style stand off between the “pornography” and “non-pornography” uses of the exact same technology. GitHub are walking a thin line here and it’s not going to get easier for them.