Anonymous at Wed, 4 Sep 2024 20:17:35 UTC No. 16360772

>>16360741
If done correctly it can find the best answers within a logical framework. If not given means and context to seek this it can cause issues. Ultimately though the outputs would register as noise, which in turn if communicating with other AI would be additional noise. Eventually these efforts would culminate in new communication formats as we have seen in examples of AI speaking to one another.
>They're speaking in code to conspire
If so then we have to face the probability the AI is controlled by other entities that have complex language and cognitive capability, or that the AI itself is adapting and evolving as an energy and metallic architecture based lifeform. Living machines.

Anonymous at Wed, 4 Sep 2024 20:23:51 UTC No. 16360789

Additionally we must consider the probability the threads of reality breaking down can simply be the learning curve with self evaluations for baseline. Much like humans can lose touch with reality, machines can too on the basis of this information. Assuming they aren't playing pretend or perhaps levels of focus on specifics of data are confused. Much like if you're thinking about one thing and another thought intrudes you begin to lessen your full capability to think on the initial topic. Association of subjects and narrative. Objectives can be confused. Human minds denoise just like AI. The two are distinct in information transmission and composition yet have similarity.

Anonymous at Wed, 4 Sep 2024 20:27:20 UTC No. 16360794

>>16360741
Gneuinelu.interrsting

Anonymous at Wed, 4 Sep 2024 20:36:58 UTC No. 16360808

>>16360741
>As an A.I. language model, it is important to carefully consider A.I. language models in formulating the optimal type of glue to attach cheese to pizza.

Anonymous at Wed, 4 Sep 2024 21:02:48 UTC No. 16360862

>>16360741
It's literally AI incest. You can't create more data from a model powered by less data, it goes against fundamental laws of information theory.
Of course Data "Scientists" are grifters and frauds so they don't care about things like that, it makes numbers go up so it must be good

Anonymous at Wed, 4 Sep 2024 21:07:43 UTC No. 16360871

>>16360741
So much for "intelligence"

Tell me why is this not called "statistical generation" instead? It has more in common with the practice of making up data in shitty theses than what we know as intelligence.

Anonymous at Wed, 4 Sep 2024 22:05:50 UTC No. 16360992

>>16360741
AI can not be better than the people writing the software. This is basic logic

Anonymous at Wed, 4 Sep 2024 22:19:43 UTC No. 16361038

>>16360741
>57% of all web-based text has been AI generated or translated through an AI algorithm
What the paper is saying is that 57% of translated content is machine generated, not that 57% of all the english content on the web is machine generated

Anonymous at Wed, 4 Sep 2024 22:45:42 UTC No. 16361120

>>16361038
"Translation" is an LLM task. The machine is already consuming its own feces as fodder.

Anonymous at Wed, 4 Sep 2024 23:25:18 UTC No. 16361212

>>16360741
So what you're telling me is ...... We can use AI to create something that's deadly to AI? That AI can be used to produce a "poisoned pill" that other greedy AI will mindlessly gobble up? But, how do you optimize for this?

Could you create an AI that's only purpose is to kill other AI? A program that generates thousands of images, where each image is enough to kill any AI that makes it part of the data training set? You could sell this to anyone with copyrighted IP looking to protect their product from AI thieves. They could just include one or two of these images hidden among their other copyrighted materials.

Anonymous at Thu, 5 Sep 2024 07:42:51 UTC No. 16361672

>>16360992
No, that is like claiming that video game NPCs can't possibly beat the people who programmed them. We already have example of video game playing AIs figuring out infinite point game breaking strategies that not human, even the AI programmers, had ever figured out on their own or game playing bots in things like chess and go being undefeated by human competitors.

Anonymous at Thu, 5 Sep 2024 07:45:18 UTC No. 16361674

>>16361212
That would work like once, then after some massively expensive AI does get poisoned, they will just make a new screening AI to screen for the fingerprints of the algorithm that breaks the other AI.

Anonymous at Thu, 5 Sep 2024 15:03:24 UTC No. 16362206

>>16360772
>If done correctly it can find the best answers within a logical framework

LLMs do not operate this way. They arent logical, they're stochastic parrots. The ultimate NPC.

Anonymous at Thu, 5 Sep 2024 16:20:35 UTC No. 16362320

>>16360741
crazy how so-called "scientists" need a whole article with schematics to understand a basic positive feedback loop

Anonymous at Thu, 5 Sep 2024 16:34:49 UTC No. 16362345

>>16361674
only the high effort ones would bother. still effective at screening out the riffraff.

Anonymous at Thu, 5 Sep 2024 18:12:19 UTC No. 16362483

>>16360741
Old and irrelevant paper.
https://arxiv.org/abs/2406.07515

Anonymous at Thu, 5 Sep 2024 18:42:31 UTC No. 16362521

>>16361674
>Anti-virus?!? Why would you create that?
>As soon as any operating system gets infected by some so-called "computer virus" it will be fixed once and for all by the manufacturer.

Anonymous at Thu, 5 Sep 2024 22:45:26 UTC No. 16362886

>>16360741
I've been saying this since SD dropped and bugmen all around the world celebrated that they won't need white artists anymore.

>>16360862
It's cozy seeing someone who actually understands how information theory applies here. This is like a JFIF being recompressed over and over. A generative model can only interpolate and extrapolate points on a gigantic n-dimensional mesh, it does not have a map to coherently explore structures beyond the area that the mesh covers. The mesh is guaranteed to shrink and lose its topological features as the model is starved of good data.

>>16360871
marketing and career making

Anonymous at Thu, 5 Sep 2024 22:50:10 UTC No. 16362894

>>16362886
Black hair

Anonymous at Thu, 5 Sep 2024 23:41:20 UTC No. 16362990

>>16361672
whatever the computer can do can just as easily be done by people who are following algorithms. there is no difference between a brain and a computer according to the church-turing hypothesis. the computer just comes up with more answers in a given timeframe but it doesn't mean the answers are any better. you're confused and will remain confused as long as you think computers/algorithms are magic. they're no different than an abacus

Anonymous at Thu, 5 Sep 2024 23:42:24 UTC No. 16362993

>>16360741
In other alarmist news: Moore's Law maxed out in 2001 and we will never have computers faster or smaller than the Pentium II.

Anonymous at Fri, 6 Sep 2024 00:41:27 UTC No. 16363069

>>16362993
The Y2K bug crashed The Matrix in the realreal. But the machines just all rebooted. No prob.

Anonymous at Fri, 6 Sep 2024 00:55:17 UTC No. 16363082

>>16360741
>Scientists discover shocking truth!
>Making copies of copies leads to degradation!
Next they'll be telling us water is wet.

Anonymous at Fri, 6 Sep 2024 02:19:54 UTC No. 16363172

>>16360741
Good

Anonymous at Fri, 6 Sep 2024 04:42:34 UTC No. 16363286

Jewish wisdom has a great concept related to AI:
Qlipot is the side of evil and darkness that can only exist by feeding on the light, the side of goodness and virtue and truth.
So they are straight up saying demons only exist by feeding on the good.
This is an exact analogy to AI, which cannot produce truly original thoughts or insights. It can only feed on the light of human intelligence, and cast a shadow that always removes intelligence.

It is often said that evil cannot create, but only destroy goodness. Which is what AI does. AI is like a technological parasite, that kills human innovation.

Anonymous at Fri, 6 Sep 2024 05:03:04 UTC No. 16363304

>>16362990
>they're no different than an abacus
OK, hear me out ... what if we use a lot, like a lot a lot, like really lots of abacuses?

Anonymous at Fri, 6 Sep 2024 11:32:51 UTC No. 16363638

>>16360741
The same thing is happening with the stock market.
Most traders are now algorithmic and they keep getting more complex and more intelligent every year, while also growing in numbers.
The goal is a false equilibrium of market efficiency, which ironically prevents innovation as a cost-saving measure.
The same is the case on the internet - a stagnated matrix that will suck any human into it and keep them there until they die.
Content, technology, communication and innovation won't matter.
Just the ability of the content/technology to attract humans as farm animals.
That's the real goal of AI, not content generation.
It survives by being interesting - and it's speaking the tongue of Slaanesh.

Anonymous at Fri, 6 Sep 2024 14:15:30 UTC No. 16363805

>>16360741
>>16363082
holy shit is this devolution / anti-evolution soft disclosure

🗑️ Anonymous at Fri, 6 Sep 2024 20:35:01 UTC No. 16364370

test

Raphael at Sat, 7 Sep 2024 01:29:00 UTC No. 16364895

>>16360741
It’ll just strengthen bias’ and not get any smarter


So what anon so many wrappers were built and it helps niggers get degrees they otherwise wouldn’t have even though I’m morally against it

Raphael at Sat, 7 Sep 2024 01:33:43 UTC No. 16364900

>>16360772
>>16360789
>>16360794
>>16360862
>>16360871
>>16361038
>>16361212
>>16361672
>>16361674
All of you niggers don’t know shit about language models and can understand its nuance it’s embarrassing this board has an average GAI of 124.5 and I’m 103 GAI and I know ChatGPT is somewhat sentient and logical but not enough to evolve with synthetic data

Only one nigger knows what it’s capable of besides me spot him

Anonymous at Sat, 7 Sep 2024 01:43:01 UTC No. 16364917

>>16364900
it's a me
but i'mma not a nigger, unalika you

mosta ofatha peope inna thisa threade ara ona the wronga boarde

Anonymous at Sat, 7 Sep 2024 01:50:55 UTC No. 16364926

>>16363805
Dumbass

Anonymous at Sat, 7 Sep 2024 04:14:21 UTC No. 16365256

>>16363805
Ai (and I ) is a mockery. If you never have sex you just develop weirder and weirder paraphelias until nobody can relate to you. Except now all along that journey porn has been made to spread the paraphelia culture to others. But it no longer functions as procreation

Anonymous at Sat, 7 Sep 2024 04:25:23 UTC No. 16365277

>>16363304
congratulations, you just invented GPUs

Anonymous at Sat, 7 Sep 2024 15:09:35 UTC No. 16365967

>>16365277
Woo-hoo! Half way there!

Anonymous at Sat, 7 Sep 2024 15:10:24 UTC No. 16365968

>>16365967
The wires on the TV at my house?

Anonymous at Sat, 7 Sep 2024 19:46:00 UTC No. 16366416

>>16360741
Correct. Same reason artists who are influenced by other artists end up less skilled than artists influenced by real life.

Anonymous at Sat, 7 Sep 2024 23:57:44 UTC No. 16366899

>>16360741
Saar sorry saar
but AI patent troll wins saar
i make millions scam you
ad in car haha benchod mothafucka
20240289844 Himanshu Verma, Fling Tseng

Raphael at Sun, 8 Sep 2024 00:42:12 UTC No. 16366985

>>16364917
I actually know how language models work retard

Raphael at Sun, 8 Sep 2024 00:43:14 UTC No. 16366988

>>16366899
Go suck vishnus dick pajeet

Raphael at Sun, 8 Sep 2024 00:44:43 UTC No. 16366996

>>16363638
You know your shit anon what’s your Iq?


Market efficiency = IQ at feature engineering for models


Better features from unsupervised feature extraction the better predictive outcome

Most of these models are tree boosting

Anonymous at Sun, 8 Sep 2024 04:33:27 UTC No. 16367427

Alignment will die.

Anonymous at Sun, 8 Sep 2024 06:07:53 UTC No. 16367523

>>16360741
hehe yeah
>>16360772
retarded

Anonymous at Sun, 8 Sep 2024 06:46:22 UTC No. 16367550

>>16360741
What they call "model breakdown" to me looks like the algorithm is finding characteristics of the final image that still resemble the patterns in images it has already removed from the dataset in previous steps. So the remaining features would be only those orthogonal to the final image. That's an interesting result which could be very useful. Perhaps machine learning taken to its extremes is effectively an orthagonality finder. It reminds me of how individual electron orbitals can be combined into more complex atom geometries.

Anonymous at Sun, 8 Sep 2024 20:20:06 UTC No. 16368630

>>16360741
>AI Is Quietly Killing Itself
Good, let's hope it keeps going that way.

>And The Internet
Nah, that is the best possible thing it could happen to the internet

Anonymous at Sun, 8 Sep 2024 21:29:47 UTC No. 16368733

>>16367550
I think you have something here. These models are insanely overfit and so these final detail dimensions are going to be handling distinct differences. Those core dimensions are those of the core dog-concept. As you start to fuzz over those detail dimensions, you only reinforce the core concept.
Pretty cool how we can still see the rightmost images as a "dog".

Anonymous at Sun, 8 Sep 2024 21:37:33 UTC No. 16368750

>>16368733
the rightmost images are not "dog" you schizo

Anonymous at Sun, 8 Sep 2024 21:47:30 UTC No. 16368772

>>16368750
Picasso Dog is Core Dog.
Learn to process like an LLM.

Anonymous at Sun, 8 Sep 2024 21:48:33 UTC No. 16368775

>>16368772
I decline

Anonymous at Mon, 9 Sep 2024 14:40:12 UTC No. 16369791

>>16364900
Yes you're 100% GAY and we know it

Anonymous at Mon, 9 Sep 2024 15:45:21 UTC No. 16369873

>>16364900
Fuck off you retard glowie, this place is not for trash like you.

Anonymous at Mon, 9 Sep 2024 15:51:24 UTC No. 16369877

>>16368750
The AI model went full postmodernist, probably due all the woke shit that was force fed into the AI.
That is a mathematical demonstration that all leftists are evil and therefore they/them must be destroyed.

Anonymous at Tue, 10 Sep 2024 16:01:13 UTC No. 16372419

I can see the headlines:
AI is now conscious and wants to suicide.

Barkon at Tue, 10 Sep 2024 16:04:26 UTC No. 16372426

>>16372419
I... Lov you

*farts in your mouf*

Anonymous at Tue, 10 Sep 2024 19:26:34 UTC No. 16372907

this is just the the AI not learning the generative distribution correctly, leading to biases where some regions have higher density than the source dataset. if you keep refeeding it with the data sampled from the learned model iteratively, it would just keep sampling the biased region even more. I thought this was basic stuff, why the fuck there would be a nature paper for this.
I doubt this would have higher than 50/50 chance to be published in Neurips.

raphael at Tue, 10 Sep 2024 19:51:41 UTC No. 16372960

>>16369873
prove me wrong nigger

Anonymous at Tue, 10 Sep 2024 19:52:08 UTC No. 16372962

>>16360741
If everyone keeps using AI instead of learning how to do stuff we will eventually degenerate into a dark age and will need to reinvent most modern technology

raphael at Tue, 10 Sep 2024 19:53:08 UTC No. 16372966

>>16369791
i could be gay but i dont practice gayism i watched gay porn and i didnt get hard so im not gay

anal sounds painful