So, brought on by this thread:
https://hexbear.net/post/8986189?scrollToComments=false
And @EdlritchEconomics@hexbear.net pointing out it's a AI image; here is a whole thing.
So I think the general consensus is AI slop bad and we dont want it in this town.
AI slop is now a bit tricky and I wouldnt blame someone for being tricked and posting slop as if it were real because they were themselves mislead. I dont blame em, scrutinizing every image is an exhausting ordeal.
So my proposal, is that any AI slop presented as true should be removed. Posting AI slop to comment on AI slop seems fine and honest. I dont think anyone posting AI slop and presenting it as real should face consequences unless there is evidence of doing it on purpose but we cant have robot pictures pretending to be real pictures. We need Blade Runners.
I dunno if a task force or an if you see something say something method is best, but it's just gonna get harder to spot for a while until the whole thing turns into a fun house mirror, so it seems worth getting in front of. You shouldn't have to scroll the comments to find out a posted image is AI and I would prefer no image posted here be AI unless it's meant to comment on that image as an AI image, illustrating a point about a slop image is one thing but if an image is AI I think it should just be removed but ill accept a tag being needed.
I picked Chat cause I dunno if anyone uses Feedback nowadays.
This also just made me think of photoshopping pictures to make them look like AI for plausible deniability/for a bit
Yeah I'm little bit tipsy so please don't ask me to elaborate but the AI upscaling on phone cameras makes me also feel like it's a continuation (deliberate or otherwise) of poisoning the well
Probably. They probably see multiple benefits from it, for example adding more things to "stuff AI does" increases their stock growth, while poisoning the well is an added benefit.
We're kinda fucked with identifying AI if laws mandating watermarks don't happen.
WeChat kinda does it, short videos made with AI are required to mark themselves as such, and content suspected to be made with AI are also flagged. From my feed:
The former has been declared by the creator, the latter has been flagged as suspected AI content. I don't know what programs these videos were made with, but both have added watermarks (含AI生成内容 'Contains material created with AI' and 内容仅为AI创意 'This content is solely an AI creation')
There are plenty of false positives and negatives, but it's something.
The problem with that is it's literally impossible, like mandating watermarking of all images touched by an image editor or of rendered CGI images. Open source local models running on open source hardware can never be watermarked because people can just make it not do that, and even slop from a compliant slop machine can just have the watermark denoised out by a non-compliant one or just be cropped for less intrusive ones.
Nah it's not. The AI companies can be forced to do it and open source local models can be forced to do it too.
There will be a fringe of people who bypass these restriction, like there is a fringe of people who shoplift. It would massively improve the problem though and it would completely eliminate AI use from creative works in entertainment, like movies or videogames.
I don't think you appreciate the global scope of AI grifter companies, like the endless parade of "unrestricted" image generation services that are just rebranding after payment processors shut them down and operating out of random countries to escape liability, renting servers and selling the outputs of whatever open source model and frontend server they feel like. Nor the degree to which the hobbyist scene will hammer away at any locked down model they get their hands on to rip out guardrails and make it spit out more gooner slop, which then just gets endlessly copied and passed around.
For that matter, we're already so far past the point of extremely strong, scary, and trivial to run models proliferating that there really isn't any chance for "well at least going forwards future things can be told to lock it down somehow and maybe someone will figure it out or something" to actually work. Things like Z-Image and Krea 2 are just horrifying, and then there's Lightricks for video which is, uh, like watching a pod person try to imitate infomercial acting but it's also still absolutely horrifying for all that it's still offputting, useless dogshit.
Death penalty for non-compliance. Drone strike them.
I am 10% joking
The more im catching up the more over 100% im feeling.