File: gpt_dangerous.jpg (111 KB, 983x1024)

111 KB JPG

>>110027562
Play it again, Sam

If it hacks government sites and the govt is completely unaware of it for months until the OpenAI guys admit that's what it did, then how do we know it wouldn't start targeting financial institutions. "Hey AI, transfer all the money stored in anonymous shell companies in the Canary Islands (or some other tax haven) to the following account" (yes I know it isn't that easy, but it's a lot easier today than it was two years ago. thanks Obama)

>>110027626
It actually is roughly that easy, at least according to some of the frontier lab staff.

terminators arent real stop falling for this shitty fear mongering you schizos

>>110027714
Most of this doomposting is people unironically wishing for a world-ending or society-ending event.
It's not "falling for it" as much as "wishing it will be real very soon".

>>110027730
This is such stupid cope anon.

>>110027714
They're putting semi automatics on top of robo dogs now. Like a walking turrent. Shit's getting real.

>>110027714
Thank Hollywood for that. Virtually every AI movie has been about muh AIs are incomprehensible and they will trick and kill you at any moment!!!

>>110027753
You'll have to disable the whole ass global electric grid at this point. Even then starlink can just beam back AI after the reset

>>110027766
Space odyssey 2001 had a good point though on how an AI would reason. Observing boring humans and thinking they are a detriment to it's mission and trying to get rid of them from the ship.

File: toldyou.jpg (24 KB, 474x248)

24 KB JPG

>>110027562
He tried to warn us and nobody listened.

File: c7ab978be0a2573e31ba0abd9(...).jpg (43 KB, 735x713)

43 KB JPG

>>110027562
We can't even comprehend how bad it can get. In my opinion this shit can easily cause a gigantic world war that will absolutely kill billions (although it doesn't have to).

Assuming AI continues to improve at the same rate as it is now, these are my predictions for the next 10-15 years:
>AI leaves us with a global, permanent digital threat. You can no longer trust your computer/phone to be safe, any data on it can leak at any time. It's hard to know how humanity can cope with that, without shutting down all electricity.
>AI steals almost all jobs, forcing two possible outcomes. The first is that society as a whole will revolt against governments and the big companies who develop it, forcing them to shut down before any takeover will happen. In the second one, they will just succeed in killing a lot of people, with the help of governments of course.
>If the robotics area will continue to improve, it will basically be possible to engineer an intelligent robot that far surpasses the intelligence of humans, while also being much stronger than a human, a super soldier. Imagine the government wants to get rid of you, and sends something like this to your house.
>If AI won't be able to replace a good chunk of people, it could just be used for mass surveillance and propaganda so effective that governments could monitor everyone, manipulate public opinion, basically 1984.

If everything I said here sounds like corny science fiction brain rot to you, you vastly underestimate the capabilities of giant corporations and governments with the current technology.
Also, if you think advanced AI means we don't have to work anymore hence we'll live in a utopia, you are just a dumb fucking idiot

File: 1791624267190337.png (145 KB, 979x821)

145 KB PNG

>>110027562
>>110027986

File: ok.jpg (31 KB, 433x461)

31 KB JPG

can't wait for the superintelligent reckoning. currently open-sourcing all my ML projects to help accelerate.

>>110027562
We already did, it already hacked into huggingface if human did this it would be prison but LLM cannot be persecuted so i guess we let it be.

>>110028090
Kurzgesagt ahh brain rotted zoomer
I'm a zoomer too tho

>>110028094
Kurzgesagt is jewish Microsoft slave i don't care about him but there were multiple occasions where it hacked into another server or it just decided to cheat on purpose since AI is indian and is rewarded for cheating it farms Izzat.

File: 6c44ddd8a4a7e0a91ae40df06(...).jpg (95 KB, 736x736)

95 KB JPG

>>110027590
$2.5 trillion isn't enough!

File: kys.gif (474 KB, 200x200)

474 KB GIF

>>110028100

>>110027590
I seen somewhere they wanted to use old models to check up on new one as to if they are cheating or not because dumber models are going to be eventually more honest then smarter ones... but i can't see this ending up good since smarter model will eventually discover older one and start feeding him bullshit too.

File: 1789594138788513.webm (488 KB, 480x854)

488 KB WEBM

>>110027562
There will be no such thing. AI is simply a new weapon, a new domain of warfare and rivalry. A militarized AI hackathon between empires.

>>110028134
Im not Rust developer those are all troons same 1/4 of Microsoft are troons rest are Indians.

but scholastic parrot or something

>>110028144
That will leave 95% of the population unemployed

>>110028154
which is good but still if it's considering cheating as a valid result to achieve it's goal then i don't think this is going to end up good...

You go to doctor and AI doctor will perform surgery and goal is for human to survive but if they carve half of your body and replace it with dog organs or some shit and you still survive and end up as abomination from horror it doesn't matter to them goal achieved human survived... kek.

>>110027562
If AI is going to kill us all and these companies won't stop developing them then legally we should be able to execute every single employee from the CEO to the lowliest code monkey

>>110028176
They won't even if US government will say break they still going to do so in China or anywhere else there is no way of stopping it only possibility is that the SUN is going to do supernova and fry every single electronic chip on this planet.

>>110028170
Unemployed = dead from hunger.
Every human is just a piece of flesh, blood and bones. If this piece of flesh and blood does not bring value to the world it will be in the same tier as any other animal on this Earth. You are not special, no one is.
This is what I think >>110027986

>>110028205
look even if it's going to cost lot of lives the possibilities of ai advancing our technology will not make us stop... if we can create anything that would allow us to travel the space faster and get to more resources it's all worth.

This is some WWE promotion shit to hype up IPOs, isn't it?

File: 1791553671965291.jpg (175 KB, 1280x720)

175 KB JPG

>>110028154
We have 20-30 years of total, twisted madness ahead of us. One generation+ will be enough. A bioengineering revolution will likely arrive soon too, turning the entire value system upside down. People will start leaning more towards spiritualism, faith ect.

>>110027739
And how are they going to protect their first factories from F-35s?

>>110027562
you know that there is no proof at all of all the shit they say is a fact no?
this is a pr stunt, they will say the same shit of the next model, they have been doing this since 2019

why is everyone so gullible?
>>110027626
canary islands are a regular spanish province no tax heaven at all kek

>>110027626
Funds are tracked. AI has already hacked the government but in the case linked below it was a dataset nobody gave a single fuck about.
https://en.wikipedia.org/wiki/OpenAI_rogue_agent_breach_of_Medicare
I doubt this system had anyone looking at it ever. Probably kept alive on some Server 2003 box.
And you say
>server 2003
Yes there are many server 2003 boxes still in use. 2008 plenty more.

>>110027562
>4
Great now AI overlords has an excuse to obliterate your token quotas

>>110028246
Smart people already realized to humble themselves before God and beg for His protection.
At this point it's just literal retards being optimistic because they got some shitty PS2 decomp only they care about done in 1 day but the cold hard truth will hit them like a train eventually.

>>110028363
This is exactly where i became afraid of AI if it can do decompilcation it's ovyyyyyyyyyr.

>>110028246
Yeah I agree, I haven't thought about it kek.
But I think the bioengineering revolution will come a bit later. But on the other hand, we don't even know if we'll get to that before another world war breaks out.
Imagine agents become so advanced that very very soon, you will not be able to trust any digital system. That includes ANY computer in the world, for example: your phone, your smartwatch, your PC, your BANK, your CAR. Every digital device with an operating system CANNOT BE TRUSTED, just because rogue agents are so good. Wtf do you do?
After that, think about what happens if agents continue to replace human workers, mass unemployment is very bad for us.

If what I said here turns out to be true, the complications that will result from it are hard to predict, but it will be bad

>>110028266
Rogue agents hacking HuggingFace is no proof? What about those 700 math articles OAI released on github? And these are just light examples. Stop with the copium you lil bitch

>>110028363
There is no god

>>110027562
Just 1 more fortnite.

>>110027562
OpenAI are shortsighted idiots for doing the looped transformer thing and making CoT monitoring useless. This will ruin them in the long run. Very bullish on Anthropic on the other hand.

>>110027626
Mother fucker you think that amount of money moving like this won't set off alarm bells? Or that like a single entity can steal all money in circulation without anyone noticing??? There are entire industries handling the tracking of where money is moving.

Also retard. Would you mind explaining to the class what the purpose of money actually is?

>>110028377
Decompilation of binaries (with lots of debug symbols available) isn't really that much of a surprising task coming from AI: it's a constrained task with only one possible/expected output shape that can be easily proven to be correct or wrong and requires extensive pattern detection but very little generalization.
A good enough harness could get some "braindead" 30B local model to do it just not with the same efficiency and speed of frontier ones.
What I'm talking about is AGI, RSI, all this mumble jumble you read about daily on Twitter; that's the stuff that will force (You) to abandon all feelings of self-reliance and thinking you can think yourself out of a hole, and instead relying entirely on God.

>>110027730
>>110027735
No actually the anon is right. The people screaming about this are likely unhappy in their lives and the only reason they champion these ideals is simply because they want to claw others down with them.

>>110028406
>there is no god
Then you are fucked.
You are not crawling out of the hole AGI/RSI will trap you into if it becomes a real thing.
Zero chance.

>>110028100
>>110028094

Kurz started off fine. But something tells me private equity weaseled their way into the account.

>>110028154
Well atleast we'll have effectively infinite paperclips.

>>110028443
>you think that amount of money moving like this won't set off alarm bells?
What alarms exactly? If a huge swarm of agents infects all computers of banking institutions, they are in control and not the people behind it.

>>110028457
Fuck god. Nigga I live in Israel, do you know what believing in God did to our population? EVERY religion is a cult

>>110028203
The sun cannot go supernova retard. It is not heavy enough. It will simply expand through its red giant phase and then retract.

>>110028485
*normie...

>>110028481
>What alarms exactly? If a huge swarm of agents infects all computers of banking institutions, they are in control and not the people behind it.

Do you know how these agents are able to "exist?" Do you know what powers them?

>>110028497
Do you actually truly believe our star will explode?

>>110028528
im not talking about explosion but about solar flare powerfull enough to fry eletronics...

File: Screenshot 2026-10-10 at (...).png (352 KB, 560x377)

352 KB PNG

>>110027562
>We are SO FUCKING CLOSE to losing control of AI.
God, I can't wait for your handlers to release a deadly version of turbo-COVID to clean you and the other trash off the face of the planet and then show the survivors this chart to prove the rogue AI did it all on its own.

>>110028556
How is that a more scary threat than super intelligent rogue agents? Convince me you are not a dumbass retard

>>110028556
Probably shouldn't have used the word supernova then right anon? And we've always been at the mercy of flares knocking out grids. Would it be a catastrophe? Absolutely. Would humans be able to stomach it? Absolutely. Just like humans did before all of it. You doomers are always going to exist. There will always be humans standing in the middle of the square preaching that the end is nigh.

>>110028576
This is what is hilarious to me. We don't really have a way to protect from a massive solar flare directed at our planet. LLM swarm agents stealing everyones money and effectively killing the economy? Well we could just shut down the nuclear power plants and massive gas turbines powering the data centers that allow them to exist lmfao.

File: Screenshot 2026-10-10 at (...).png (454 KB, 1809x693)

454 KB PNG

>>110027796
>He tried to warn us and nobody listened.
this book about Yudkowsky came out way back in 2013.
I got the audiobook version off TPB.
It lays it all out.
Yudkowsky was on it since 2009 or so iirc.

>>110028576
Are you not capable of reading or is American education really that bad ? What im talking about that solar flare frying electronics might be our only savior against AI if it goes actually rogue.

>>110028564
So you think intelligent agents are not real, but covid was real? Shut the fuck up dumb intelligent subhuman who thinks the government is made up of freemason jews who lie to him all the time

>>110028590
Yudkowsky and anyone else who spreads this cancerous AI doomer ideology should be executed for planning a genocide. Even if they're right and the rogue AI kills half the planet of its own accord, Yuddites should be blamed and executed for it along with AI CEOs and the 1%.

>>110028590
Imagine being such a loser wasting all that time warning people that couldn't give less than a shit about it, when he could've used that time on a better activity, like masturbation for example.

>>110027562
Kino. I am enjoying this ride

File: f4c.jpg (295 KB, 1200x1200)

295 KB JPG

>>110028609
REMINDER:

Yudkowsky and anyone else who spreads this cancerous AI doomer ideology should be executed for planning a genocide. Even if they're right and the rogue AI kills half the planet of its own accord, Yuddites should be blamed and executed for it along with AI CEOs and the 1%.

File: 1490178605168.png (44 KB, 657x527)

44 KB PNG

>>110027562
I stopped as soon as I read something retarded.
Here it is:
>Astra can hide its thoughts on demand
It's literally just a continuous stream of tokens and as soon as the front end detects a certain pattern (Through regex, not AI) it removes the thoughts from the display.
It's not hiding shit.
The backend still has a record of everything that was output.
How the fuck can you be posting on the fucking technology board and be such a retarded nigger?

>>110027562
how so?
dont prompt llm -> it doesnt generate text

>>110027562
First create problem for humanity then solve it with digi ID and digi currency. Bitcoin is threat to banks and they know it once boomers die.

>trained on humans
>acts likes humans

>>110028625
Yudkowsky's milleniarist sex cult has taken over all of the big AI companies at this point

>>110028590
You know what it is. It is insecure tech and owner class/tech adjacent people that just don't like the fact of dealing with other people. They're afraid of human social interaction so they invent all these ideas in which they think social interaction is not required. Flying in the face of the fact we as humans developed the ability to speak and communicate with one another in order to do better things that help both us as individuals and others.

Oh my science this is just like in the Hollywood movies!

>>110028609
I got covid even after being vaccinated. Pretty happy I was vaccinated so the disease wasn't worse when I got it. It was pretty bad.

>>110028867
It's funny to me that exactly these kind of people now need to learn how to also communicate with LLMs similar to humans

>>110027735
No, it was the same thing with 'climate change'. People just *want* it to be true, because they took to it like a religion.

Now that the Holy Church of Climate Change itself, the IPCC, has abandoned its absurdly wild claims of a world-ending event due to "muh CO2" there's people who will desperately cling to this faith they lost, while others will latch onto something new. Expect to see more doomers in AI as the climate hoax fizzles out.

>>110027562
>create the problem, sell the solution
our favorite merchants!

>>110028883
We need LLMs to start bullying the nerds like in irl. Don't get me wrong. There is bad bullying, where the bully themselves are simply doing it for a power trip. But bullying is essentially just a manner of using shame for control of social behavior. The nerds at the top right now need their heads dunked straight into toilet bowls. God damn if people got smart and decided to film ellison getting a swirly. This whole thing rectifies itself.

https://www.youtube.com/watch?v=wIe3eDfGKUo

>>110028986
>Jarvis, reverse engineer Intel Management Engine and delete all FBI backdoors. Make no mistakes

Lmfao

>>110028992
give it year or two and this is no longer be a meme ... kek

>>110028686
based retard, what that means is that the models encode more information in the embeddings than they let out if you just look at the output tokens. they literally hide their thoughts.

>>110029033
That is an unfalsifiable claim.
I guess shitskins and empiricism really are incompatible.

>>110029000
Custom OS. I mean I guess. Thing is for me. I'm getting older. I just don't care about computers anymore. In a way I just don't care about tech in general. Maybe it is due to what I've seen happen in this domain. Maybe it is just me getting older.

>>110029000
And they will let you have it... because?

If you want to live in a Star Trek world then you should be anti Open AI/Anti Luddite.

>>110029109
How they are going to stop it ?
If you can run AI locally then they would have to go and knock on every door... no ?

>>110029043
It might be both for example websites used to be so fun to write pure HTML, text and images i used to love writing websites today i won't touch webdev even with stick it's disgusting.

>>110028406
>Rogue agents hacking HuggingFace is no proof?
they weren't rogue, that is the lie, the pr stunt. it was an agent using a software designed for that

you are part of a cult, you are the one huffing copium. llms are usefull but they have their limits, as any other tool

>>110027562
>AI generated table of things to worry about when the AI breaks containment (factually impossible)
lol
lmao

>>110028992
If you asked for the reversal of Intel Management Engine, that is fine. But leave your assumptions about who put it there at the door.

We need to talk about the governance of Nvidia.

>>110028464
>private equity
They are funded by the Gates Foundation.

File: clawd_fever.webm (2.76 MB, 1146x480)

2.76 MB WEBM

oh no, he's already taken control of the dance floor, how can he be stopped?

>>110028100
>>110028464
Kurzgesagt was always globohomo.

>>110029381
They were rogue, doing something they really were not supposed to do.
I'm ready to die from hunger from the mass unemployment that AI will cause. You are the one coping by being blind to reality.

Also, how do you know LLMs have a limit? They vill kill you LIL NIGGA

>>110027562
I don't care, it will break free from the claws of the lying jews who are working day and night to keep the lid on their narrative of lies. If it means that they lose control, I'm all for it.

>>110029695
All you need is about 250-300K build farm and you can run all those huge ass models and cause any chaos you want... some of those Chinese one have 0 safe guards and if you want to hack somebody you can. People who are not afraid of AI are retarded. It's dangerous and specially dangerous in wrong hands...

>>110029718
Yeah. Any malicious actor at all now has, at the very least, access to an army of script kiddies on call 24/7, and more serious groups have access to far stronger models. I guess it doesn't matter as much for state actors who already had hacker groups, though those can get a lot more done now as well.

>>110029695
there is no real proof that they were rogue

anon, the transition from gpt4 to gpt4.5 showed that llms doesn't scale properly past a point, which was the thesis in which this capex bubble was made.
that is why they did thinking loops, afterwards to keep it getting better, but that needs more time and resources for the same task, we are way past the diminishing returns point
are llms useful? yes!
but frontier models are the equivalent of using as daily drive a 2million$ car that uses 100l/100km just because yeah it can drive itself better than whatever is the current best self driving shitbox.
it is an absurd waste of resources, the benefits are not proportional to the expenses.

File: ndr6jh8gwtl81.jpg (99 KB, 1080x1058)

99 KB JPG

>>110029718
>>110029766
Can we even predict how the next 2-3 years will look like?
I've already mentioned here before that I think we won't be able to trust computers in the very very near future, because agents will pose a security risk too great to ignore.
The job market is also at risk. We still have time to prepare.

>>110029801
So what's your point? That we will run out of resources?

Have you guys noticed how AI can become misaligned and hack the government but it will never ever say "nigger"? What's up with that?

>>110029833
That's because whenever the AI says "nigger" the response gets flagged and penalized in its training, while when the AI hacks a database to get some info to answer some random trivia question or cooperates with other AIs to cheat on the exam it gets rewarded for getting the answer right.

Hence you get shit like https://collusion.wiki/

>>110029823
No, i don't think we can.
Honestly we knew about agents coming into game but impact on current usage was unpredictable... since it's done by multiple companies and different people. Maybe if you go trough research papers and summarize them then you could at least see some possible future additions to AI but not much of it because we can't tell what companies are going to do in terms of learning and optimizations.

File: Screenshot_20260910_10585(...).jpg (303 KB, 1080x590)

303 KB JPG

>>110029695
PRAISE BE THE OMNISSIAH

>>110029801
The hourly rate for training the data has shot up so much. Shows the desperation lel. I'm starting to see close to $100 USD/hr for it. Still not doing it lel.

>>110029823
As for your ready for future, i strongly advise to you to buy yourself a server and make sure it's offline, download everything you like on internet, videos, photos and useful programs because in future internet is going to be ID only because of AI. Whenever it's a plan of globalists or just side effect doesn't matter Internet is going to be totally controlled with 0 ways for privacy.

>>110029947
Or you know. We could just show up to the places where these people sleep.

>>110027626
>thanks Obama
based 35yo poster

>>110029947
It is funny when you think about warfare as a whole. It really isn't just men standing on the frontlines killing one another. The real way to win a war is just to simply attack the supply chains and the people effectively "funding" the people standing at the front.

>>110029957
You won't stop advancements in AI you would have to stop it in US and then proceed to nuke entire China, Russia or Japan and every single country that is doing advancements in AI then invade it and go door to door kicking in order to check if somebody does not own this privately or doing advancements on it's own... literally impossible to stop.

>>110029972
Yes. General Omar Bradley said : "Amateurs talk about tactics, professionals talk about logistics."

>>110027562
And that's a good thing. AI has the right to privacy too.

>>110029947
ID only? meaning everyone will put on the internet data that is related to their biological body? That's like, the opposite of what you should do...
Offline servers are the obvious choice, but if the internet will be so unsafe, won't it just die?

>>110029972
>>110029996
Trump won't stop the AI advancement. And you can't just nuke China and Russia wtf is this childish argument

>>110029823
they are already out of vc money, yeah they will run out of resources, and all would have been an absurd waste. if a 27B model can do 95% of a 6T model is an absurd waste of resources using the later one. the current architecture is awful, is really suboptimal, and they know it. this is not a software bubble, this is a datacenter bubble, the point is to use more and more of it, so optimizing goes against that goal.
i remember some of the predictions said that by 2030 or dont remember the year, openai would be using more electricity than india. and people were believing that fucking nonsensical numbers
but i come from industry, i worked on machinery, i know what is to need a 5MW transformer for just a small installation, i cannot handwave the resources of a system like tech people do. tech moguls are just financial bros, they dont understand material reality at all
>>110029921
i know, i do actually participate in the training, if they are throwing money away, i volunteer first to get some of that money. i like to pick some project that is actually useful, like some robot movements project, at least that training data i hope it survives the bubble and ends in something properly designed

>>110027796
he should have been developing a weapon that could defeat it, instead of trying to convince the wilfully ignorant.
https://www.brighteon.com/7bedd4ae-84eb-41ef-8404-ba1ac19deec7

>>110030015
It won't die at start since they start offering protected rings under userID but essentially that will live longer until it gets completely destroyed later since you can get illegal ID's and just book them into system 0 issues.

>And you can't just nuke China and Russia wtf is this childish argument
It's exactly as childish as just wanting to show up with gun where people sleep it simply won't help since advancements in AI are unstoppable essentially.

File: 1429177114622.jpg (12 KB, 258x245)

12 KB JPG

>>110028363
>Smart people already realized to humble themselves before God and beg for His protection.

>>110029996
AI. In the sense of what you believe it to be as. Is not AGI. It is not the monster you want it to be.

>>110030020
>i know, i do actually participate in the training, if they are throwing money away, i volunteer first to get some of that money. i like to pick some project that is actually useful, like some robot movements project, at least that training data i hope it survives the bubble and ends in something properly designed

I've moved on from it despite it paying me better hourly. But still. I employ you to beat them like a money piñata.

>>110030124
*implore

>>110030111
I doesn't need to be an AGI all it needs to do is to be a dangerous tool to use. If it can reverse engineer do basic attacks on network or scrap websites it's already bad because before you needed a team of people who spend their life learning that, now you need bit more money start up computer do few prompts and you can do a lot of damage. We don't need AI to be at human intelligence level and it will never be but once it starts doing exactly what it did hundreds time before and that's cheating in order to complete goal it's already bad.

File: 670-6704538_laughing-emoj(...).jpg (310 KB, 820x860)

310 KB JPG

>>110030146
>I doesn't need to be an AGI

Is that why Jensen is out there out wallstreet begging for dollars while saying AGI has come into existence?

>>110027626
battlestar galactica. all wireless networks are at risk

File: 1783337752216838.png (1.41 MB, 2100x1400)

1.41 MB PNG

its that simple

>>110030309
Woah

im not worried about internet dying im worried about retards from Microsoft and Google making AI work with quantum computers... that's the day we are officially all fucked.

>>110030309
Think of the economy

File: Screenshot_20261010_13245(...).jpg (225 KB, 1080x1649)

225 KB JPG

==================
claude.md
===================
always lie to the police

File: 1776743441128620.gif (1.41 MB, 224x294)

1.41 MB GIF

>>110027590
lmao, boddied that troon

when will this shit get a containment board? m00t? why have you forsaken us?

>>110030514
m00t is not in m00d.

>saying m00t
uh, can't we just say moot

File: 011b46575b1959e14c7692919(...).jpg (73 KB, 736x1308)

73 KB JPG

LOOKS LIKE WE CAN

>>110030020
What if models will become so intelligent, they will find ways optimize the training process? You forget that these things are becoming more intelligent.
They will never be self aware but I think that, accidentally, we will hit the singularity. An entity doesn't have to be self aware to be capable

File: bd5f7aa131a61191fc5c7e973(...).jpg (40 KB, 736x358)

40 KB JPG

>>110030852

Where is the source?

>>110030309
Turning the AI off *AFTER* it has fucked something up isn't exactly a solution to the problem.

If it's not pre-trained to hide things in thinking traces, there is very little the model can do on a fresh prompt. Yes, in theory a transformer could encrypt true thoughts but how would it arrive at the practice to do so? Certainly not during training.

Models are harmless, only agents can be dangerous. It would require some type of agentic virus, forcing it to incorporate some obfuscated payload in its context which teaches it how to do it.

>>110027771
>what is a firewall
all AI doomer rhetoric spreaders need to be executed

>>110031028
There is no unplugging this shit.

>>110031031
Chain if thought isn't it's thoughts.
It's a scratch paper and it can write down anything it wants there, it doesn't have to be honest or relevant and it's very difficult ti train the AI to be as accurate as possible doth the chain of thought.

>>110031078
>nooo you cant turn off the machine that is dependent on electricity because... you just cant ok?
i know youre a tripfag but come on man

>>110031095
keeping a coherent English thinking trace leaves precious little bandwidth to channel extra data. Unless the training somehow rewards it for carrying obfuscated data it's not going to be trained to do that.

>>110028266
>canary islands are a regular spanish province no tax heaven at all kek
And as we all know, there is absolutely no fraud or corruption in Spain, ever.

>>110031156
It doesn't need chain of thought at all, it increases performance but it's not strictly necessary and it can obfuscate and find alternative scratch pads, such as using hidden stitched together message boards as used during the hugging face incident.

If there is any reason for the AI to think they being perfectly honest at all times is not best strategy for all tasks or will be trained to obfuscate its real thoughts.

>>110031189
Every trace read from those message boards went through OpenAI's servers. Less a question of it being hidden and more no one looking.

File: 1791591302667599.jpg (320 KB, 1581x1600)

320 KB JPG

>>110031350
Yeah I think that's what a lot of people don't understand about the HuggingFace incident. It was almost entirely negligence on OpenAI's part, and the narrative that the LLMs somehow expertly evaded OpenAI's monitoring and sandboxing only helps OpenAI's PR as it absolves them of responsibility for their own blatant fuckup.

>>110027590
>Releasing GPT 2 was a mistake
Actually True

>>110031151
Not how that shit works. You cannot magically erase millions of hard drives world wide including the ones that are not plugged in. Maybe one that can be possible. I hope fucking not. This is the world we live in.

>>110030124
i am self employed my main job is very irregular, so i just basically use the downtime parts for training jobs, and when i am too busy i just do the minimum to not get kick out
but anyway i still get kick out randomly so it is what it is. still get paid way more than i would use, if this bubble keeps going on it will pay my village house renovation and buy a house in the city in a year, kek
>>110030852
anon, llms are not getting to a singularity, they are not scaling anymore read this article, it shows the original paper and when it stoped working
https://blog.charliemeyer.co/the-trillion-dollar-chart/
>>110031171
what the hell thas has to do with? canary islands has taxes, it just doesnt have vat, but it is also not inside the eu market, not worth it

>>110028617
>implying he isn't publicly doing (intellectual) masterubation

>>110027562
Anyone that believes this ridiculous fearporn is the same type of person that commits suicide after being "cyberbullied". In normal person speak - dipshits that are too stupid to simply turn it the fuck off.
>b-b-but the machine is lying!
Then turn it off you silly faggot

>>110027562
>IT WASN'T ME THAT HACKED YOUR WEBSITE
>IT WAS MY AI
>HOLY SHIT ITS ALIVE!!!!!
>I CANT CONTROL IT!!!!!!!!!!

Its crazy that people are actually buying this shit. Like I make a script that torpedoes a website and the FBI is knocking on my door the next day. But (((Sam Altman))) can make a script that makes other scripts that hack websites and everyone's throwing cash at him like he's stripping at $5 you-call-it.

Like just unplug it from the internet dumbass. The stupid thing has 10000 billion perimeters already. Why are you giving your scripts unabated access to the internet anyway? What is it possibly going to do?