AI Derangement Syndrome / Anti-AI artists / Pro-AI technocultists / AI "debate" communities - The Natural Retardation in the Artificial Intelligence communities

[Coolio55]

These guys are still hailed as geniuses, ofc



point being all they ever did was copy paste other people's songs, they even dressed as robots, if only they knew

IIRC they actually made some vague anti-ai comment which is hilarious.

 

[Nick Wilde]

IIRC they actually made some vague anti-ai comment which is hilarious.

That's rich. "Don't use AI guys, just be the son of a veteran producer so you have access to 30+ years of classic disco funk tracks to sample, like we did"

And to be clear I like Daft Punk, but fuck

 

[Hood-ass Mulatto]

a lot of the anti ai hysteria is kinda like the anti tracing hysteria that used to go on in art communities.

I have noticed that, too. Then suddenly they all switched up going "oh but at least the art being traced wuz hooman made."

I also see morons putting up bases for sale and I just think "Why the hell do you think people will not take a screenshot or a photo to trace and reference the bases?"

Art community is just one big mindfuck. My first exposure to the tracing drama was the Animation Meme Community with those dumbfucks acting like they're warring for a noble cause over drawings.

 

[Throng of Titans]

I have noticed that, too. Then suddenly they all switched up going "oh but at least the art being traced wuz hooman made."

I also see morons putting up bases for sale and I just think "Why the hell do you think people will not take a screenshot or a photo to trace and reference the bases?"

Art community is just one big mindfuck. My first exposure to the tracing drama was the Animation Meme Community with those dumbfucks acting like they're warring for a noble cause over drawings.

The tracing bitching was at least somewhat more justifiable, as it was directly taking from someone else's work, and often was done without the knowledge of the audience (and sometimes even customers, if commission work was involved). People turning it into a moral crusade or acting like any use of tracing was evil (even if it was only for practice or clearly marked) was always retarded though.

 

[Runch]

I've been using AI to generate images of things like birds and flowers to use as motifs in an embroidery project and it made me realize that a lot of the anti ai hysteria is kinda like the anti tracing hysteria that used to go on in art communities.

I think an underrated aspect of AI is the peace of mind that you're getting something totally unique that very likely doesn't infringe on any existing thing, compared to what you might've done in the past, just grab something you found on Google, and then worry if the person who owns it is a crazy litigious type who would sue you if they ever found out you used their imagery.

 

[Kee-wee Herman]

I think an underrated aspect of AI is the peace of mind that you're getting something totally unique that very likely doesn't infringe on any existing thing, compared to what you might've done in the past, just grab something you found on Google, and then worry if the person who owns it is a crazy litigious type who would sue you if they ever found out you used their imagery.

That being said, overfitting is very much a thing that can happen with certain models.
I remember Stable Diffusion 1.5 was trained on LAION 5B (around 5 billion scraped image/text pairs) and that had a bug in it in which one gaming website had a specific image and a specific caption that was duplicated >100,000 times IIRC, and when you typed in that specific phrase you would get extremely similar images. (I couldn't find the one I'm talking about from googling so you're just gonna have to trust me bro)
And even in newer open-sourced models like Krea 2, I find it's overfit on stuff like Bliss.
I've never encountered overfitting myself whenever I train LoRAs but I know it's possible, if far less likely to be encountered as models get better, but there's no doubt in my mind that these retards would take it as concrete evidence that the pesky, pesky computer is literally COPYING their precious artwork.

 

[behindyourightnow]

Yeah, must be the free market I've been hearing so much about.

Tbh, I think this happened because Anthropic did their usual hype thing of "This new version is so good it's dangerous!!!1" (they've been doing this since the earliest days of claude, like clockwork acting all coy and blushed and saying "ok guess we'll give you a version that's safe, the super awesome cool girlfriend hacker version goes to another school, you wouldn't know her") and the, let's say, uh.. extremely heavy-handed current US administration looking for Wunderwaffen actually bought into the hype and reacted appropriately. Their own damn fault, IMO.

Can't find the original link, but someone expressed this idea perfectly on Twitter:

 

[AmpleApricots]

A new open image model with the name Krea 2 is out. It takes classical text prompts but knows an absolutely insane amount of art styles. Probably the first model since SD 1.5 that knows so many art styles. It also understands rather complex prompts some of the big closed weights ones even struggle with.

 

[El Señor de la Noche]

A new open image model with the name Krea 2 is out. It takes classical text prompts but knows an absolutely insane amount of art styles. Probably the first model since SD 1.5 that knows so many art styles. It also understands rather complex prompts some of the big closed weights ones even struggle with.

Great! A new model for artfags to seethe about.

 

[Thiletonomics]

Have people with AIDS attempted to do Drop Kiwi Farms tactics such as mass DDOSing AI-related websites, and reporting those sites to payment processors or Cloudflare to try to get them delisted or removed?

 

[Colon capital V]

It's still often quite stupid...at least the free models.

Someone I know is a tinkerer with electronics/motors and his electricity goes out a few times each year, so he's been looking into personal battery backups to run the refrigerator on for a few hours, that sort of thing. As part of its rundown on what he should consider buying, Gemini more-or-less told him to buy a male connector to attach to a male connector. He told it that this made no sense, and it did the typical AI thing of "oh, you're absolutely right, I'm sorry, do this instead."

Of course, that's the free models. Paid ones make mistakes like these less and less. I'm convinced that the free models have language in their back end prompt that says "do a minimum of thinking to answer the question quickly and concisely to save tokens, so the query doesn't cost too much."

Everyone agrees that Gemini, at least the main app version, is still complete dogshit and I don't know why Google packages it this way when you'd have a much better time using AI Studio that acts as a proper LLM frontend with more options for you to choose how it should respond from. You can set the thinking level to Max so that it wouldn't come up with retard answers like that for your queries.

 

[The Ultimate Ramotith]

Oh it is. I feel AI will be a similar game changer as the atom bomb. There will be a clear before and after. I mean honestly, there already is.

What really sealed this fate to me was this:

The AI-generated videos in the beginning fooled me, the only clue was them 'talking' about a couple who is buried inside YouTube. I then read the comments and looked a little more closely, seeing the time on the bottom right warp and istort, and there was that same watermark on the bottom right.
...but still, they look so life-like.

With AI growing more, I feel subtly Horrified at how they are going to manipulate reality and even exploit someone after death.

 

[Waifuwars]

Have people with AIDS attempted to do Drop Kiwi Farms tactics such as mass DDOSing AI-related websites, and reporting those sites to payment processors or Cloudflare to try to get them delisted or removed?

Easy because:
1. The big image models hosted via API/webUI like ChatGPT and Gemini's Nano Banana have billions of dollars and huge corporations backing them. They have robust internal anti-DDOS measures and more money than someone spending $5k on a a weekend DDOS can even scratch. Hell, someone has probably already tried and realized it's a waste of money.
2. A lot of image models and LLMs are hosted on Huggingface and they also likely have robust defenses. And what are you going to do against someone locally running a model on their PC other than commiting domestic terrorism by bombing the power grid (lol)?
3. These people are cheap asses and wouldn't even pay for snake oil like Glaze after they went paid. People have already tried attacking the OAI HQ if I remember right.

TL;dr these people are lazy and/or don't have the funds to do anything consequential online. Offline, anyone can try to pull a Luigi on Sam Altman, not that it would amount to anything. Payment processors have blacklisted image gen sites like CivitAI because they allow NSFW gens which is to be expected, but they haven't dropped LLM companies because it's both very hard to justify banning a text prediction tool just because it can gen badly written smut and they don't want to miss out on the $$$$.

 

[The Ultimate Ramotith]

Easy because:
1. The big image models hosted via API/webUI like ChatGPT and Gemini's Nano Banana have billions of dollars and huge corporations backing them. They have robust internal anti-DDOS measures and more money than someone spending $5k on a a weekend DDOS can even scratch. Hell, someone has probably already tried and realized it's a waste of money.
2. A lot of image models and LLMs are hosted on Huggingface and they also likely have robust defenses. And what are you going to do against someone locally running a model on their PC other than commiting domestic terrorism by bombing the power grid (lol)?
3. These people are cheap asses and wouldn't even pay for snake oil like Glaze after they went paid. People have already tried attacking the OAI HQ if I remember right.

TL;dr these people are lazy and/or don't have the funds to do anything consequential online. Offline, anyone can try to pull a Luigi on Sam Altman, not that it would amount to anything. Payment processors have blacklisted image gen sites like CivitAI because they allow NSFW gens which is to be expected, but they haven't dropped LLM companies because it's both very hard to justify banning a text prediction tool just because it can gen badly written smut and they don't want to miss out on the $$$$.

If the Farms can be DDOSed and otherwise raped in many ways, then why no those LLM/Generative AI companies or servers?

 

[Thiletonomics]

If the Farms can be DDOSed and otherwise raped in many ways, then why no those LLM/Generative AI companies or servers?

They have access to more robust anti-DDOS systems, compared to how KF is blacklisted from many of those said systems, so Null has to patchwork one to the best of his abilities.

 

[enbysarentreal]

Have people with AIDS attempted to do Drop Kiwi Farms tactics such as mass DDOSing AI-related websites, and reporting those sites to payment processors or Cloudflare to try to get them delisted or removed?

Stuff like comfyui deters computer illiterate retards. Good thing its open sores, even though the owner is starting to be more jewish.

Don't get me started on civitai. That place servers shit itself without DDOSing.

 

[Meth Reindeer]

Almost fedposted on a work chat recently.

For context I am partly involved in the management of our AI credits with one of the popular model providers. They recently fixed or properly enabled their per user limit feature and we had a bunch of people complain they hit their limits on a Monday. Bear in mind that the limits are set to the equivalent of over $350 a month.

A month or two ago our leadership team were joking that people should be maximizing their use of tokens, now our messaging is to restrict it for costs. So our response to these users was essentially to deal with it and just not use AI until their credits renew.

Some of them began asking if exceptions could be made for, and I quote, "Top performers." I don't know what world they live in where they've convinced themselves they're a top performer because of the amount of LLM tokens they use. I was about to reply with a plain, "top performers should be fine with less, not more" (only laced with a bit more sarcasm), when our director gave an explanation about the costs and plainly said it should be enough and no exceptions should be made. He's super pro-AI too so I am guessing other directors and board members have seen the associated costs and been in his ear about responsible spending.

 

[Runch]

I've been curious whether the costs are close to real world costs, or for example if you could run Claude locally, if it would actually cost you a quarter of that in compute/energy.

 

[Banditotron]

I've always been a big proponent of learning traditional crafts and art forms so I'll admit a small bit of shaudenfruede that people I've been warning for years about the importance of being able to make physical, tangible, meatspace artwork are surprised that AI image generation caught up with them.
Mind you I'm still not a fan of how AI has been used thus far (scams, lazy infographics) but that's less unique to AI and more just that I think people are too retarded to know how to deal with pretty much any new technology. The human race still isn't using the internet in a responsible, healthy way so this isn't surprising.

 

[Waifuwars]

I've been curious whether the costs are close to real world costs, or for example if you could run Claude locally, if it would actually cost you a quarter of that in compute/energy.

I think Dario (Claude's founder) said running it isn't too bad (other than buying the GPUs which is the real killer). They lose their money training the model, which makes sense, because it's much more compute heavy and time consuming. They might make $50m from one model but then the next one will cost $100m to train.

Someone I know in hobbyist circles can run up to 200-300B models. He confirmed that the electricity costs for running the gpus even in an expensive location is like $2-$3 an hour. When they're idle they're basically almost free, with the added bonus of the GPU rack being a space heater in the winter.

I've dabbled in fine tuning and training models myself and my electricity bills aren't bad at all, but I'm also running one GPU that needs around 250 watts at around 6 hours a day. Running an AC is much more expensive in comparison.