this post was submitted on 25 Oct 2023
115 points (100.0% liked)
Technology
37551 readers
46 users here now
A nice place to discuss rumors, happenings, innovations, and challenges in the technology sphere. We also welcome discussions on the intersections of technology and society. If it’s technological news or discussion of technology, it probably belongs here.
Remember the overriding ethos on Beehaw: Be(e) Nice. Each user you encounter here is a person, and should be treated with kindness (even if they’re wrong, or use a Linux distro you don’t like). Personal attacks will not be tolerated.
Subcommunities on Beehaw:
This community's icon was made by Aaron Schneider, under the CC-BY-NC-SA 4.0 license.
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Many "AI generated" images are actually very close to individual images from their training data so it's debatable how much difference there is between looking at a generated image and just looking at an image from its training data in some cases at least.
That's a symptom of overfitting, which requires the image to be repeated in the training hundreds or even thousands of times. That generally only happened in earlier image generation models, more "modern" ones ("modern" in this case being measured in months because this is such a fast-developing technology) have much better curation of their training sets to avoid exactly that sort of thing. Nobody wants AI image generators that replicate images from their training sets, what would be the point?
So if you want to find an image model that gives you a close duplicate of an existing image of child abuse, you'll need to find one that was sloppily trained with a training set that included hundreds of duplicates of child abuse imagery. I kind of doubt you'll be able to find one of those.