You think these billion-dollar companies keep hyper-illegal images around, just to train their hideously expensive models to do the things they do not want those models to do?
Like combining unrelated concepts isn’t the whole fucking point?
True enough - but fortunately, there’s approximately zero such images readily-available on public websites, for obvious reasons. There certainly is not some well-labeled training set on par with all the images of Shrek.
Yes and they’ve been proven to do so. Meta (Facebook) recently made the news for pirating a bunch of ebooks to train its AI.
Anna’s Archive, a site associated with training AI, recently scraped some 99.9% of Spotify songs. They say at some point they will make torrents so the common people can download it, but for now they’re using it to teach AI to copy music. (Note: Spotify uses lower quality than other music currently available, so AA will offer nothing new if/when they ever do release these torrents.)
So, yes, that is exactly what they’re doing. They are training their models on all the data, not just all the legal data.
It’s big fucking news when those datasets contain, like, three JPEGs. Because even one such JPEG is an event where the FBI shows up and blasts the entire hard drive into shrapnel.
Y’all insisting there’s gotta be some clearly-labeled archive with a shitload of the most illegal images imaginable, in order for the robot that combines concepts to combine the concept of “child” and the concept of “naked,” are not taking yourselves seriously. You’re just shuffling cards to bolster a kneejerk feeling.
It literally can’t combine unrelated concepts though. Not too long ago there was the issue where one (Dall-E?) couldn’t make a picture of a full glass of wine because every glass of wine it had been trained on was half full, because that’s generally how we prefer to photograph wine. It has no concept of “full” the way actual intelligences do, so it couldn’t connect the dots. It had to be trained on actual full glasses of wine to gain the ability to produce them itself.
I’m saying it can’t combine clothed children and naked adults to make naked children. It doesn’t know what “naked” means. It can’t imagine what something might look like. It can only make naked children if it has been trained on them directly.
You are not logged in. However you can subscribe from another Fediverse account, for example Lemmy or Mastodon. To do this, paste the following into the search field of your instance: [email protected]
Video game news oriented community. No NanoUFO is not a bot :)
Posts.
News oriented content (general reviews, previews or retrospectives allowed).
Broad discussion posts (preferably not only about a specific game).
No humor/memes etc…
No affiliate links
No advertising.
No clickbait, editorialized, sensational titles. State the game in question in the title. No all caps.
No self promotion.
No duplicate posts, newer post will be deleted unless there is more discussion in one of the posts.
No politics.
Comments.
No personal attacks.
Obey instance rules.
No low effort comments(one or two words, emoji etc…)
Please use spoiler tags for spoilers.
My goal is just to have a community where people can go and see what new game news is out for the day and comment on it.
You think these billion-dollar companies keep hyper-illegal images around, just to train their hideously expensive models to do the things they do not want those models to do?
Like combining unrelated concepts isn’t the whole fucking point?
No, I think these billion dollar companies are incredibly sloppy about curating the content they steal to train their systems on.
True enough - but fortunately, there’s approximately zero such images readily-available on public websites, for obvious reasons. There certainly is not some well-labeled training set on par with all the images of Shrek.
Yes and they’ve been proven to do so. Meta (Facebook) recently made the news for pirating a bunch of ebooks to train its AI.
Anna’s Archive, a site associated with training AI, recently scraped some 99.9% of Spotify songs. They say at some point they will make torrents so the common people can download it, but for now they’re using it to teach AI to copy music. (Note: Spotify uses lower quality than other music currently available, so AA will offer nothing new if/when they ever do release these torrents.)
So, yes, that is exactly what they’re doing. They are training their models on all the data, not just all the legal data.
It’s big fucking news when those datasets contain, like, three JPEGs. Because even one such JPEG is an event where the FBI shows up and blasts the entire hard drive into shrapnel.
Y’all insisting there’s gotta be some clearly-labeled archive with a shitload of the most illegal images imaginable, in order for the robot that combines concepts to combine the concept of “child” and the concept of “naked,” are not taking yourselves seriously. You’re just shuffling cards to bolster a kneejerk feeling.
It literally can’t combine unrelated concepts though. Not too long ago there was the issue where one (Dall-E?) couldn’t make a picture of a full glass of wine because every glass of wine it had been trained on was half full, because that’s generally how we prefer to photograph wine. It has no concept of “full” the way actual intelligences do, so it couldn’t connect the dots. It had to be trained on actual full glasses of wine to gain the ability to produce them itself.
And you think it’s short on images of fully naked women?
I’m saying it can’t combine clothed children and naked adults to make naked children. It doesn’t know what “naked” means. It can’t imagine what something might look like. It can only make naked children if it has been trained on them directly.
Incorrect.