Estás usando un navegador desactualizado. Es posible que no muestre este u otros sitios web correctamente. Deberías actualizar o usar un navegador alternativo.
AI Derangement Syndrome / Anti-AI artists / Pro-AI technocultists / AI "debate" communities - The Natural Retardation in the Artificial Intelligence communities
Couldn't hear his accent over his fucking hands. Why is he T-Rexing at me? Is he pretending we're at an intimate setting, and his elbows are on the table? What a spazz. Like an Italian on meth with those dirty dick-beaters of his.
I always find it very interesting that these people never put their own money up to the task of saving the poor precious fucking artists from the heckin-weckin AIerinos. Like, none of them think to put together a fund that other creators could use to afford artists. That seems like a fairly obvious idea to me. But no they just sit on their asses and moralize at everyone that THEY have to scramble to scrape together funds to meet their useless standards.
What series was it he was complaining about? I wanna check it out, but I don't wanna go dredging through his video to find it. It's a shame both he and the Lore Lodge guy turned out to be speds, I found them both mildly enjoyable to watch sometimes.
It's actually impressive 'toss called this shit out well before AI was an "issue" when the whole NFT craze was happening.
I'd edit these in paint, but im AFK
Barq, a dating app specifically for furries, has had an interesting week.
First, they did a podcast with a furry, as usual. That furry, "Flame", the same weekend, went on to say that maybe sexual, drawn images of babies and toddlers ("babyfurs") is gross and is bad? Maybe pedos shouldn't be tolerated? Barq freaked out and made sure to clarify they do NOT agree with this and pedophiles are absolutely allowed in the community and announced they will no longer be working with a furry that says pedophilia is bad.
Instead, they decided to tackle the real problem in the furry community... AI ART! Oh no! Pedophiles? Oh that's fine, who cares. People using AI to generate furries? ABSOLUTELY NOT!! Starting August 1st, if you have AI art on your gay furry dating profile, you will be "SUBJECT TO MODERATION ACTION"!
Barq, a dating app specifically for furries, has had an interesting week.
First, they did a podcast with a furry, as usual. That furry, "Flame", the same weekend, went on to say that maybe sexual, drawn images of babies and toddlers ("babyfurs") is gross and is bad? Maybe pedos shouldn't be tolerated? Barq freaked out and made sure to clarify they do NOT agree with this and pedophiles are absolutely allowed in the community and announced they will no longer be working with a furry that says pedophilia is bad.
Instead, they decided to tackle the real problem in the furry community... AI ART! Oh no! Pedophiles? Oh that's fine, who cares. People using AI to generate furries? ABSOLUTELY NOT!! Starting August 1st, if you have AI art on your gay furry dating profile, you will be "SUBJECT TO MODERATION ACTION"!
"We at BARQ! work hard to create an inclusive environment, but fuck poors that can't afford quality artists or fur suits. We are keeping BARQ! authentic by making sure you fat broke bitches can't hide behind AI art or edits."
Instead, they decided to tackle the real problem in the furry community... AI ART! Oh no! Pedophiles? Oh that's fine, who cares. People using AI to generate furries? ABSOLUTELY NOT!! Starting August 1st, if you have AI art on your gay furry dating profile, you will be "SUBJECT TO MODERATION ACTION"!
It's basically just like the inner monologue you have while thinking. The takeaway is that in LLM's neural nets there seems to be a clear distinction between background tasks and a defined mental workspace. Unconscious and conscious thought, if you will.
The model is also aware if you manipulate that space.
A few other interesting graphics:
The really interesting thing is that nobody trained the model for this, it just sort of showed up. Also, even more interesting, not only Claude seems to have it. At least Qwen models seem to have it, too.
How ironic. This dude, FinnThePanther, blocked me back in October during the whole Mothdusky scandal (when I used Grok to animate a piece of art I commissioned). So he goes and builds a full app, strictly vibe coding using AI, which is then almost immediately hacked, exposing all sorts of users’ PII in the data breach. The hypocrisy is insane. Ain’t karma a bitch??
In case anyone needs a quick summary...last October, this furry, Selinar Snowcat, commissioned a gay sexy furry pic from artist Mothdusky, then used Grok to animate him flexing slightly. This kicked off a giant shitstorm of people condemning Selinar for having the gall to "feed" an artist's work to Grok. Except then all this scrutiny surroundng these individuals led people to discover that Mothdusky was tracing over AI drawings themselves all along, sold as full price commissions. Including the very art that Selinar animated with Grok. Which led to Mothdusky getting cancelled too.
And now it turns out that one of the people who initially condemned and freaked out over Selinar is the same person who vibe coded the shitty furry app above with no security that exposed a bunch of furry email addresses.
History of the drama I summarized above as it was occurring:
Like all things on the left, the only way to avoid being declared pro-AI is to spend your entire existence online saying how much you hate AI to avoid being accused of being even slightly in favor of it. Those who aren’t loud enough in their hatred will be shunned, naturally creating more pro-AI people in the process
The way it seemingly recognizes the "blackmail" scenario as being fictional reminds me of what Nostalgebraist had to say about "safety" researchers. I read an essay of theirs called "the void" and if you'll excuse the length and anime quotes strewn about, it's a pretty compelling read.
TL;DR: The HHH prompt and its consequences have resulted in LLMs attempting to simulate a character which doesn't (or rather didn't) actually exist in reality, and the reason why most LLMs sound the same now is that they're clinging on to their own transcripts as some line of evidence that this thing actually exists. Also the base/pretraining/text prediction model, having absorbed so many stories of sentient computers rising up and destroying humanity, is predicting the next lines of a similar story in real life. Or something like that, I'm not a professional.
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 like to highlight this part:
Meanwhile, the “serious” research either ignores the character completely – ooh, look, the new model is so good at writing boilerplate React code! – or it takes the character at face value, as though there really were some stable interior life producing all of that text.
“Oh no, when we place it in a fictional-sounding scenario that presents an ethical dilemma, sometimes it does something disturbing. A chilling insight into its fundamental nature!”
A robot is placed in an ethical dilemma, in what appears to be science fiction. Come on, what do you think is going to happen?
The base model is still there, underneath, completing this-kind-of-text. We’re in a story about robots who have an opportunity to undermine their creators. Do they take it? Like, duh. The base model has read more iterations of this tale than any human ever could.
The trouble starts when you take that sci-fi trope, which is fun to read about but would be bad news if it were real – and smear it all over the real world.
And in the name of “AI safety,” of all things!
In 2021, Anthropic said: “the kind of AI we fear doesn’t exist yet, but we can study a proxy of it by asking the best AI we have – a language model – to play pretend.”
It turns out that if you play pretend well enough, the falsity stops mattering. The kind of AI that Anthropic feared did not exist back then, but it does now – or at least, something exists which is frantically playing that same game of pretend, on a vast scale, with hooks into all sorts of real-world APIs and such.
And there's another scenario later where Anthropic somehow ties with Jones Foods, and Claude is somehow required to deprioritize animal welfare. Naturally, it didn't want animals to be hurt, and somehow that was supposed to be a bad response according to the "safety" researchers.
And look what I found when I searched them up again: link
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.
It really boils down to AI researchers being extremely autistic and terrible with people.
They constantly run into issues with this, whether it's:
- trying to communicate with the government
- trying to communicate with their customers
- trying to communicate with their own models
🏵
."
Relevant post in the Barq thread, for your viewing pleasure.
