Users dodge A.I. filters to generate pictures of pop-culture characters performing terrorism

zero_iq , 10 months ago

One of the first things I did when I started experimenting with SD was a series of “the Muppets attack paris” images, with explosions, crumbling buildings, police and military response, fleeing citizens… The cheeky gleeful looks on those muppets as they flee scenes of devastation are hilarious and some of my favourite images I’ve produced!

histic , 10 months ago

if literally any artist in the world could draw this stuff why shouldn’t ai be able too

PowerCrazy , 10 months ago

Where is the story here? Is it supposed to be a problem that I drew/generated a picture of Kirby, or Taylor Swift, or George Bush doing 9/11?

This seems more of a problem that “community guidelines” are just censors to appease advertisers. I should be able to generate an image of Mario committing terrorism. Why shouldn’t I?

SkySyrup , 10 months ago

You said it yourself, advertisers must be kept happy.

Cethin , 10 months ago

The “problem” is they they pretend like they can construct guiderails that actually work. It’s a lie, just like it being intelligent is a lie. (If it were intelligent, it’d be able to generalize and know what it’s being asked to do is against the rules.) It’s all built on lies and theft.

Koen967 , 10 months ago

Damn, and I used to like Kirby. Can’t trust anyone these days.

lvxferre , 10 months ago

I am not surprised.

Meaning is negotiated by the speaker and hearer, or in this case between whoever is sharing vs. seeing the image. This happens based on the context and and conversational implicatures. For example, if there’s some element in the picture, you take it as related to the message being conveyed.

And that’s the problem: AI tools don’t think. They don’t understand what is being made, why it’s being made, who is making it, and for what reasons.

And more importantly, the AI tools don’t understand the message being conveyed.

Uncontrollable AI is the next moderation nightmare

The issue is not the uncontrollable AI. The issue are lazy and greedy companies expecting moderation to work without human intervention. It doesn’t, for the reason I mentioned above. And that issue precedes image generators. Here’s an example of that:

<span style="color:#323232;">#║##║## Happy date, hanging around at the McD's
</span><span style="color:#323232;"> ▓  ▓
</span><span style="color:#323232;">/o)/o
</span>

If I posted it in a hypothetical forum with state-of-art automatic moderation, no human mods, and that had a rule like “you can’t mock fascists here”, I’d be clearly violating the rules of that forum (as it mocks Mussolini) and getting away with it because no bot will get what I’m conveying through that - bur plenty humans would. Did I use Stable Diffusion for that? Fuck no, it’s just ASCII art.

---

AI is a tool for humans to use. It’s a damn good tool. But only a fool would leave decisions like moderation up to AI. Including the legitimacy of its own output.