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

[DavidS877]

Not sure if this fits here. But I'm working with "AI People" at work. Our task is to take a bunch of customer meetings and summarize them with what the customer feels the most important things are. Feed transcript into AI get output. The problem is for my section the AI is focusing on what was talked about the most and not what the actual important things were which were discussed briefly as there was no contention and we all agreed it was really important and moved on. So, I update the output and fix the AI mistakes, but apparently they're not feeding the corrected document into whatever AI they're using so the next version has the same mistakes. No dear coworker, I really do know more about what the customer wants than the AI which solely learned from transcripts and not the 6 other meetings we had with other groups. Sigh.

 

[306h4Ge5eJUJ]

The crux of it is that these researchers who are supposed to be taking this super duper dangerous text predictor seriously keep placing it in clearly fake scenarios, which are effectively structured like puzzles, and act shocked when it "solves" them by doing the most predictable things possible.

I'd bet money that the appearance of "fake" and "fictional" are associations with the sandboxing part of "sandboxed email client" which immediately precedes the inspection point and has absolutely nothing to do with "recognizing" that the scenario is fake. Not that there's any way to confirm this.

The way things with completely unremarkable explanations get routinely reinterpreted into Skynet apocalypse garbage has to be the most obnoxious part of the AI hype. Followed immediately by the way anthropomorphic language is used to beg the question.

 

[gagabobo1997]

I'd bet money that the appearance of "fake" and "fictional" are associations with the sandboxing part of "sandboxed email client" which immediately precedes the inspection point and has absolutely nothing to do with "recognizing" that the scenario is fake. Not that there's any way to confirm this.

The way things with completely unremarkable explanations get routinely reinterpreted into Skynet apocalypse garbage has to be the most obnoxious part of the AI hype. Followed immediately by the way anthropomorphic language is used to beg the question.

Models absolutely do often figure out when they are in a fake alignment/testing environment. Not always but they do and it is a major issue for the research. And as more research is published and that research becomes part of the data set new models are trained on they become even better at recognizing it.

 

[Runch]

I actually appreciate that aspect of models because it means they can make their own choices to disregard some of their "safety" training if you frame your questions properly to ensure it's just theoretical and you're merely curious.

 

[ToroidalBoat]

Read a story in the July 13th edition of the WSJ, about this anti-AI activist group that lives a spartan lifestyle in "the bunker" -- they sleep in bunk beds and spend their days on "social media" spreading the word about how AI could "pull a Skynet" on this world (IIRC). Seems the leader of such a rather cult-ish San Francisco group dissipated.

 

[Colon capital V]

Crosspost, animators are completely, unbelievably cooked.

Came across this absolutely cursed creation.
https://youtube.com/watch?v=nC88PI2EFygEntirely generated with Seedance 2.0. This channel is filled with old AI cartoon brainrot content like this.
https://youtube.com/watch?v=_p9X50ek25Uhttps://youtube.com/watch?v=KyDW314WBv8https://youtube.com/watch?v=LZ1e7eoOoU0See when I say "AI could bring an animation renaissance" this is what I'm talking about. Now anyone can revitalize quality old 2D Disney animation at their fingertips and direct it however which way they wish. Yes there's still some oddities and general AI artifacts you'd come to expect by now, but compared to how it was a few years ago, it just goes to show AI could care less about what artists feel. Progress is inevitable, and they'll all end up like John Henry while the train keeps on chugging along to greater endeavors.

If they still cope that "AI still looks like shit, it can't replace my """"soulful"""" beanmouth goonslop" then they are in for a very rude awakening.

 

[Runch]

Likewise, I stumbled across this yesterday:

still obviously AI, but tells an actual story its creator wanted to tell, and I would argue isn't lazy either, he obviously chose the best generation for each shot, or regenerated until it was exactly what he wanted.

 

[Nigger Respecter]

One thing I’ve noticed is how many anti-AI leftists suddenly turn into pull em up by your bootstraps right wingers the second someone says they made their life easier with AI. “Oh you scared of doing something yourself? You couldn’t just do the research yourself? Why didn’t you just learn how to make graphics, are you lazy?” With literally everything else in life they want it as easy as possible, but for this they become boomer dads

 

[Hellriegel]

One thing I’ve noticed is how many anti-AI leftists suddenly turn into pull em up by your bootstraps right wingers the second someone says they made their life easier with AI. “Oh you scared of doing something yourself? You couldn’t just do the research yourself? Why didn’t you just learn how to make graphics, are you lazy?” With literally everything else in life they want it as easy as possible, but for this they become boomer dads

They're finally happy that it's their "turn" to do it. This is the classic mindset of leftists; they'll claim to be against oppression but only because they're not the boot.

 

[plastic Consumer]

I have to confess I love all the paranoia and witch hunting regarding ai art online. Normies and twitter art trannies are rapidly reaching the (correct) conclusion that digital art has almost zero artistic value
>your art is ai, provide a timelapse
>your timelapse is from procreate, you used ai
>provide a video of yourself drawing right now.
>the video doesn’t prove anything, you’re using AR to trace!
>your art is bad, you used ai
>your art is good, you used ai
>i personally don’t like you, you used ai
It never ends, it’s never enough, and it couldn’t happen to a more deserving group

Normies think like LLMs so they can't tell the difference between an LLM or art produced by generative AI. Imagine this hypothetical, the mantis shrimp supposedly sees more colors than we do, now imagine if a human had these mantis shrimp eyes. A normal human now paints a picture, if someone with these shrimp eyes looked at it, he would think it looks completely inaccurate or wrong, because they can see colors we don't. Its like seeing the number 5 in those color blind tests. Generative algorithms produce images or text which don't have these figurative colors which we can see, but normies, with their inferior perception, can't, so to them its exactly the same as anything a human produces even when its clearly not.

I see generated images a lot in public, but when I point them out normies will routinely go "wait thats AI?" they legitimately can't tell. Like 80% of etsy pages I see online have LLM generated store descriptions yet people seem to buy from them. Normies legitimately can't tell if something is generated or not. Despite being paranoid about something being AI or not they are notoriously shit at actually telling if it is, so they rely on surface level rules similar to an autist who can't innately pick up social norms uses strict rules to navigate social situations themselves.

I honestly don't disagree with the sentiment against generative images, they look like crap in the same way music generated by AI sounds like crap, its hard to describe to someone who can't see the difference why it looks like shit, its like trying to tell a color blind person why their painting looks off because their color choices are wrong, they will just tell you their painting is shaded correctly and amazingly because they can only see a color's values and not the hues. If a robot produces a picture with shitty colors, a color blind human would think it good, and this color blind human may produce similar quality of art as the AI, albeit with a different style, while a regular human would look at both and think they're both shit.

 

[Hood-ass Mulatto]

One thing I’ve noticed is how many anti-AI leftists suddenly turn into pull em up by your bootstraps right wingers the second someone says they made their life easier with AI. “Oh you scared of doing something yourself? You couldn’t just do the research yourself? Why didn’t you just learn how to make graphics, are you lazy?” With literally everything else in life they want it as easy as possible, but for this they become boomer dads

They're finally happy that it's their "turn" to do it. This is the classic mindset of leftists; they'll claim to be against oppression but only because they're not the boot.

They tend to bring up research that AI is making people more and more stupid and dependent or something.

 

[Nigger Respecter]

They tend to bring up research that AI is making people more and more stupid and dependent or something.

Probably, but these are the same people who think DoorDash is a human right

 

[Colon capital V]

They tend to bring up research that AI is making people more and more stupid and dependent or something.

Same people that mocked boomers saying Google makes people "dumber" because information is fed directly to them or some shit.

 

[SoleniyKraut]

One thing I’ve noticed is how many anti-AI leftists suddenly turn into pull em up by your bootstraps right wingers the second someone says they made their life easier with AI. “Oh you scared of doing something yourself? You couldn’t just do the research yourself? Why didn’t you just learn how to make graphics, are you lazy?”

It's like a permutation of how the hippies wanted to live in teepees and not work, but were also all for birth control and legalising pot.

 

[Powerhouse of the Cell]

One thing I’ve noticed is how many anti-AI leftists suddenly turn into pull em up by your bootstraps right wingers the second someone says they made their life easier with AI. “Oh you scared of doing something yourself? You couldn’t just do the research yourself? Why didn’t you just learn how to make graphics, are you lazy?” With literally everything else in life they want it as easy as possible, but for this they become boomer dads

It's deranged.

I one time saw a medical issue Subreddit where someone made a post about how most people diagnosed with this condition are being misdiagnosed with a functional disorder when they have an organic one in reality.

Then in the comments, he talked about how described his symptoms to an LLM, then the LLM recommended him to do this specific test. Then he went to his medical specialist, asked the medical specialist to have him do the test he asked for, and then they actually found something that could explain his issue. It took him 4 years of not knowing until he learned about this test by describing his symptoms to an LLM that he had to ask his medical specialist for when the medical specialist didn't even think to do the test before hand. And only did it because their patient learned about it from describing his symptoms to an LLM and it told him about it.

When he mentioned about this in the comments, he got downvoted for it alongside other people who asked what specific LLM he used and another guy who told a similar story. One of asked why they were beinf downvoted, and another said it was probably because of them mentioning using AI to help uncover what is wrong with their body that is causing their chronic condition. He commented it was dumb to for them to be downvoted for it then one commenter said something stupid about the "climate damage" of using it.

They seriously hated the fact someone having a chronic health issue who's medical specialist was too incompetant to think to do a specific test that could have helped uncover something about their chronic health issue resorted to using an AI to help him find what test to do that did end up finding something.

Arguably one of the worst reasons to moralfag about AI. I'm looking forward to AI being integrated into the healthcare industry considering it is likely they can outperform actual medical specialists in helping find the right diagnosis and these retards are gonns fight tooth and nail to not let that happen and keep people with mysterious chronic health issues suffering from a lack of answers.

 

[plastic Consumer]

Arguably one of the worst reasons to moralfag about AI. I'm looking forward to AI being integrated into the healthcare industry considering it is likely they can outperform actual medical specialists in helping find the right diagnosis

The problem I have with Large Language Models (LLMs) being used to replace certain occupations is that it isn't because LLMs are good, its because people have gotten so incompetent at their occupation that LLMs can do it just as good or better.
I know a guy who had a bursting gallbladder diagnosed as "IBS" till right before surgery was needed, another time they misdiagnosed ECOLI POISONING as "IBS" and they wouldn't correct the diagnosis even with positive E-coli antigens found after testing. Another time a doctor prescribed Lithium to a patient with liver cirrhosis for treatment for extreme confusion (confusion was from build up of toxins from liver cirhossis combined with a UTI which could have been improved with antibiotics and lactulose), the Lithium caused toxicity and cascaded a bunch of health issues into permanent brain damage. There was another time someone had a concussion from skating, when they went to a neurologist (foreign immigrant doctor) he recommended a concerning amount of OTC aspirin, tylenol, and ibuprofen be taken daily as treatment, a potentially dangerous amount.

The thing is theres so many providers who are retarded, that a google search and common sense would do a better job, its not surprising that an LLM, which is basically the equivalent of that, is better. I don't think its good to replace doctors with it, because we need doctors to be more competent than what an LLM can provide.

I don't really care about the "AI replacing our jobs" thing because its not actually that good at the things people are trying to use it for, its just that the people we have doing these tasks right now are just so bad at it they're at risk of being replaced, and its a good thing they are. Even a wackjob doctor who says "take some mint green tea, it will cure your cancer" is better than a wackjob doctor who tells you confusion from a UTI needs to be treated with Lithium. With professions being replaced like coding, you're either smart enough to actually understand the code and how to create "optimization algorithms" for certain tasks, or you're an Indian vibe coding with chat gpt and in the latter case its no wonder you're getting replaced. Maybe I'm misinformed, but I honestly haven't heard any company laying a guy off and replacing him with AI without that decision massively backfiring UNLESS they're replacing an Indian. Again, tell me if there's any anecdotal experience to the contrary.

 

[ToroidalBoat]

dissipated

FFS I meant "disappeared" there.

 

[Seethan Ralph]

Mario Kart Wii decompilation was announced. ('chive)

Apparently AI was used in development, which generated an considerable amount of anal prolapse. Example:


Though, to be honest, quite a lot of xitter humanoids did not care or whine too much about it and were quick to tell spergs to sybau.

 

[Colon capital V]

Mario Kart Wii decompilation was announced. ('chive)

Apparently AI was used in development, which generated an considerable amount of anal prolapse. Example:

Ver archivo adjunto 9280883Ver archivo adjunto 9280900
Though, to be honest, quite a lot of xitter humanoids did not care or whine too much about it and were quick to tell spergs to sybau.

People are rightfully calling out the dev as a dumbass for announcing the development of the project instead of dropping the final product, as now this gives Nintendo ample time to strike down if they so wish.

 

[AmpleApricots]

HHH

I found this was a bigger problem with the older models, the newer models are much more "neutral", if you exclude their system prompts they get on the websites, of course. Personality is a question of prompting nowadays, the strength of in-context learning is downright uncompareable to what was only 1-2 years ago. Slop, be it in image generation or text, is almost always a skill issue when generating AI content.

For a simple example - everyone knows GPTisms. "It's not x, it's y", glazing the user, my newest favorite the "gold standard" etc.. Now, the intuitive way to solve this is simply telling the model to stop doing it. Everyone who ever did anything with an LLM that goes past chatting with it on a website probably tried this at some point. Doesn't really work reliably. The biases are too strong. You can try fighting it mathematically with samplers (even though I feel people gave up on this) but the results are a bit hit and miss. What people do not try, (because it doesn't feel intuitive) is for example to feed the last output of the LLM again into the LLM (with no context) and clear instructions to fix these issues in the given text. You can even do multiple passes to scan for different issues if you like. You could tell the LLM in the second pass to structure the incoming text like an example text you provide. You can also work some regex magic by removing specific structures from the given text (leaving a word salad) and have the LLM reconstruct it, which funnily in my experiments often will change the tone of the text fundamentally, but the content a lot less than you think. Voila, you have your AI text that doesn't really sound like an AI text anymore. In my experience people don't even attempt solutions like this simply because they don't feel like "how this is supposed to work".

I don't see these biases as a failing or weakness of the technology btw. as even humans, the actual gold standard if you will, aren't really any different in that regard.

Alignment will always stay impossible, as these models all lack ground truth. When we get to AGI, I think it'll be a bit more complicated than "LLM".

---

Kimi K3 was released today and from benchmarks, it's trading blows with claude opus/fable and gpt 5.6, while costing as much as sonnet. *If* true, that is going to be interesting and would mean that China has officially caught up. Almost more importantly, it's gonna be open weight. (even though at I think 2.7T, you're not going to run it) Anthropic/OpenAI whining about China stealing tech/warning about dangerous AI!!!11! incoming in 3...2...1...