this post was submitted on 30 Aug 2026
173 points (98.9% liked)
Technology
87742 readers
4259 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 3 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
What do you think it was trained on to be able to create child porn? Yes, children were harmed and will continue to be harmed to “improve the model”
The model knows what kids looks like and it knows what porn looks like. It is now more complex than the thought experiment, “the Chinese Room” because it can interpolate between two things it knows rather than basic request and response.
And use of the model drives demand for it, just like purchasing real CSAM. Even if the subject is now 50 years old, it incentivizes abusers to make new material from people that are currently underage.
Were they also trained on the vast quantity of media of gymnasts made out of pasta for them to be able to create videos of that?
One of the main points of GenAI is the ability to smush multiple concepts together. There are plenty of training data of children that are in swimwear, or just literally naked that has nothing to do with CP, and there's boatloads naked adults and of course, porn, it can successfully combine them together even if there was no actual CP in there whatsoever.
And given the internet's proclivity toward legal "teen/jailbait" porn where the models are of age but focus heavily on child-like features/poses/clothing/etc, there's literally millions of high-resolution photos that an inference machine could easily stitch together a CSAM image using nothing but legal source images.
What? Even if you're right that CSAM was included in training data (and that's highly debated), it would absolutely have been inadvertent and due to the Internet dragnet style of training. I don't know where you get the idea that AI companies would ask for more CSAM and drive up demand to "improve" their model.
That’s the thing… it was derived from real CP.