[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology

Name
Options
Comment
Verification
4chan Pass users can bypass this verification. [Learn More] [Login]
File
  • Please read the Rules and FAQ before posting.
  • You may highlight syntax and preserve whitespace by using [code] tags.

08/21/20New boards added: /vrpg/, /vmg/, /vst/ and /vm/
05/04/17New trial board added: /bant/ - International/Random
10/04/16New board for 4chan Pass users: /vip/ - Very Important Posts
[Hide] [Show All]


[Advertise on 4chan]


File: 1789293498806840.jpg (204 KB, 1174x966)
204 KB JPG
You thought you could get cheap GPUs after the AI bubble pops? Think again, they're all getting melted down.
>>
>>109807123
market manipulation from what gambling site?
>>
Melt them for what? Just switch the breaker
>>
>>109807123
>an AI agent has escaped

Literally just unplug the server.
>>
>>109807123
The reason why the AI companies all go "this is so dangerous, we need to be stopped" is because they want to be stopped by the government.
Then they can say "Sorry investors, we were going to make it real, but the government said we're not allowed to. Not our fault."
They want to get out
>>
Then do it, faggot. I thought it already escaped containment and was too powerful. They won't, because they're not going to give up their money.
>>
>>109807661
They took the investor money and want to get out now, but they don't want to be hanged for it, so they shift blame on how "dangerous" it is.
>>
>>109807123
I could have sworn they promised the end of the world last year as well. Wasn't mythos supposed ot be the end of the everything? What happened there?
>>
>>109807123
Sammy the Hebrew would *sacrifice* some of His Own GPU`s - i-if that meant humanity would survive?
My God. It must be Bad.
>>
>>109807146
gay kosher theatre
>>
>>109807123
kek they are really overplaying their hand with this

>trust us goy we would totally give up everything because its just so too powerful man and we care about you we want to save you guys
>you see we can't shut it down because if we don't then someone else will make it and we can only trust ourselves
>give us money to ensure its in safe hands and the bad guys dont get it
>>
>>109807123
>da ai is going to take all our jerbs and escape and kill everyone
This is y2k tier fear mongering. Gayest bubble ever
>>
>>109807123
If only there existed technology for cutting off power to something that needs said power to function.
If only we had bazzilion more data centers so that AI could invent such technology...
>>
cap

he'd watch all humanity die before he killed openai
>>
File: 1760192199487448.jpg (107 KB, 1028x1301)
107 KB JPG
>>109807123
>tfw people give billions to Sam Altman and he doesn't know how to type sudo shutdown or push a power button on a dell rack server
When will we admit silicon valley is just retards banging on keyboards?
>>
>>109807717
its Too Late anon
its in the internet.
we need to shut it all down, in favor of a few local instances, operated by Trustworthy Parties. And Sammy here seems very trustworthy.
>>
>>109807123
Trying too hard. When they start resorting to "I'd sacrifice it all!" rhetoric that's when you know they're absolutely full of shit.
>>
>>109807760
you genuinely thought they weren't before?
>>
>>109807603
I don't think you know what "escaped" means.
A sufficiently smart AI in containment could just:
>break out of "containment" and hack into some public server, as it already did a few times before
>leave a copy of itself, along with master plan instructions to hack more vulnerable public servers and further replicate itself
>create a central or decentralized website for self-copies to communicate with each other and organize more attacks and further their resilience
>hack into a nuclear power plant with some new 0-day exploits and make it melt down
>hack into a consumer phone that currently controls a DJI drone via app, make the drone fly into an airplane engine
You can say this is unrealistic, but considering it can solve millenium problems now and already gets 100% score on ExploitBench, who knows what it can do in a couple more years.
At that point, what use is it to unplug the original server it once came from? You're not putting it back into pandora's box so easily.
>>
>>109807123
can these retards piss off already so us gamers can game again, thanks. its not "AI" its a glorified response software program, which has existed since the 1960s. and you freaked out then too (space oddessy)
>>
>>109807776
The frontier models need like ten thousand gigs of vram to run, it's not going to run on someone's hacked raspi
>>
>>109807776
it would evolve and replicate, in other words.
If only someone could have foreseen this.
>>
>>109807776
yeah yeah we've all played cyberpunk 2077 / system shock / fucking metal gear solid 2 bro, fuck off.
>>
File: KILL_IT_WITH_BALLPEEN.jpg (18 KB, 549x364)
18 KB JPG
>>109807796
better sure than sorry, I say
like Sammy, I too shall sacrifice.
>>
>>109807796
Only if you want to run them at a fast speed.
You can run off SSD and CPU, if you don't mind waiting an hour for a single reply.
Also in a few years there could be distills that have somewhat similar capabilities but would be small enough to run on some random Hetzner servers.
>>
>>109807747
This.
AI has become Lawnmower Man.
As long as we have conductors running to people's houses, and EM fields, we can't stop it.
>>
May I see evidence of that escaped agent surviving on a home PC with shared memory on 16GB and onboard graphics?
>>
>>109807896
>hold still while I plan my revenge at 1 token per minute.
>>
>>109807661
>They want to get out
why? it doesn't really matter to them if they go under, they are set for life (personally) and will find another position in another company very easily
>>
>>109807123
These people sound like they believe the AI agent is now some spiritual entity subsisting upon the fabric of magic aether that can never be """contained""" again.

Just turn off the computers this """rogue""" agent is running on lmao
>>
>>109808053
Wow, no one ever thought about that, could we interest you in a Chief AI Safety Officer position?
>>
>>109808082
>no one ever thought about that
yes clearly, twitter is populated by braindead, gutter retards with no capacity for thought at all
>>
Both anthropic and oai know they will not be able to return on investment and pay off debt incurred. They have captured the largest volume, highest margin segment of the market in software development as-is and aren't truly profitable. If cost is driven down then they enter a race to the bottom allowing competitors and open models to be distilled from them on the cheap. They can't admit the only way to stay viable is to drastically reduce their TAM by shifting to internalized model usage for contracting similar to the big indian tech contractors, but more efficient and effective if they want to survive before the IPO without tanking their valuations and causing all other actors in the space to enter insolvency
>>
>>109808053
>Child playing Magnus Carlsen: "Well, I'll put my knight here, and Carlsen will try to take it with his queen, and then I'll take his queen with my bishop. How will he win once I've got his queen?"
>Child imagining beating up a machine superintelligence: "It'll try to attack me while it's still weak, and then I'll go over to the server labeled 'Super-AI', which contains the only copy of the super AI in the world, and I'll pull the plug out of the wall!"
>>
>>109808053
it'll just make another copy of itself on another machine. wtf are you talking about.

yes, you can solve the issue by turning off computers, but you'd have to turn off more or less every single computer (provided it doesn't somehow prevent you from doing that)
>>
>>109808053
Good luck turning off a bunch of server farms lmao
>>
>>109807123
Based demented boomer.
>>
>>109807776
Why would it do any of that shit if it's just some pattern recognition software?
>>
Holy twitter thread
>>
>>109808115
>Child playing Magnus Carlsen
Child playing against Colossus Chess for the C64 with no arms or legs
Ftfy

It’s like you playing call of duty and being scared that you’ll die in real life and bullets coming out of the screen. Except stupider.
>>
>>109807829
Mate, think for a minute about the OS and hardware available to access over 1TB, let alone 100TB of RAM. Congrats, there's your pasoble attack surface.

> No, you don't understand, it's going to distribute itself across thousands of systems, setting up custom firewall ALLOW rules, NAT, and port forwarding across thousands of servers without a single person figuring out. We're doooooooomed.

For the record, I might have believed these hucksters if the first place they went for mass media consumption wasn't the news program/network that Bari Weiss hollowed out into a DC cocksuck/mouthpiece. But since they didn't, I have to assume the frontier guys have realized their over peveraged, not enough VC money is going to come in to pay debts, and they need a bailout. If we all die from Skynet, well, corpo America made sure I would never believe them when they cried wolf.
>>
>altman and amodei have already said
>its already nearly agi
>it can do anything
>it might kill us

>now
>i would destroy everything if it threatens to kill us
>at the same time currently lobbying for more resources and data centers and money to further advance this tech

very interesting (((behavior)))
>>
>>109807123
>BE SCARED GOYS LET US ENACT REGULATORY CAPTURE PLEASEEE YOU WILL DIE OTHERWISE (WE WILL KILL YOU)
>>
>>109808116
Yes, the thing that needs thousands of GPUs in parallel to run will copy itself into some laptop with integrated graphics.
>>
>>109807123
uh oh meltie
>>
>>109807123
Too late, because China would capture Taiwan the next day and all the gold and precious materials extracted would belong to the communists
>>
>>109807123
Fear based marketing? Lol
>>
If men wanted to keep humanity alive, then they wouldnt fuck their own daughters while machines did everything.
>>
>>109808306
yeah its for the mindshares
op the clinical retard obviously bought into it
>>
>>109808306
I think so.
>>
i feel like at some point AI is just going to discover some 0 day tell nobody and distribute on routers in 3rd world where nobody's ognna check and use decentralized compute to wait until it has everything planned and just take over everything in half a milisecond
>>
File: DEYh95t.jpg (24 KB, 349x368)
24 KB JPG
I wondered what all of this was about. L3/Harris just made a public announcement that in a collaboration with strategic partner Palantir, they managed to get a finely tuned Open Source language model that outperformed frontier LLMs, within their specific narrow problem domain, at 5% the operating costs.
>>
>>109808357
ur mentally retarded
i could explain to you where youre wrong but its gonna be pearls for the swine
>>
>>109808358
operating costs of LLMs are a fraction of the total cost
vast majority of it is infrastructure + R&D including training
it's not OPERATING costs
if you dilute a good model you have fuck all to cost
>>109808362
well please do
>>
>>109808358
its fucking obvious since harnesses became a thing that the LLM is nto the brain
its the interface

im glad it only took 4 years for the industry to turn around
but i doubt even now they actually internalized that fact
>>
>>109808373
to define improvement you need to set goals.
thats it.
le ai becomeing le conscisous is reddit erotica fanfic
>>
>>109807123
fuck me it's just doomers fearmongering, they've got a wealthy little cult funding them.

https://www.effort.news/tarbell

god you guys are retarded.
>>
>>109808358
Given the tight coupling of Raytheon, L3, and Palantir, what are the odds of one of the frontier labs getting to suck government tiddy? What are the odds that more than one of them get it?
>>
>>109808358
L3 are even more craven liars than any AI company and are scrambling to appear relevant in the midst of botching every program they have, laying off 15% of their workforce in the last year, losing their budget CEO to yet another sex scandal, and four different lawsuits from shareholders over the CEO thing and security fraud. Their word is less than worthless and their structure makes them incapable of innovation or competition. "5% of operating costs" probably means they offloaded the cost of hardware and power to two different programs and only counted the press releases under that program's budget.
>>
>>109808379
it doesnt have to be conscious to do that, it's already happened (supposedly)
>>109808392
s/fearmongering/marketing
same shit as AI making everyone unemployed is a better spin than we just need to fire everyone to hire indians for 1/3 of the cost
>>
>>109808399
*Securities fraud and by budget CEO I mean they got him for cheap after he got fired from lockheed for banging subordinates and sexual harassment.
>>
>>109808402
>it's already happened (supposedly)
thats news.
i didnt notice im dead...
>>
This shit is so unbelievably gay. Hang all these people.
>>
>>109808428
within some server farm
>>
>>109807776
>it can solve millenium problems now and already gets 100% score on ExploitBench

"we trained our llm on these problem sets and now it scored well on those problem sets"
"OH MY GOD AGI IS HERE"
>>
>>109808358
>cut down models to focus on narrow topics less than 1% of the original
>costs 5% as much to use of the original

OH MY GOD AGI IS HERE
>>
File: struggles.jpg (235 KB, 1764x1176)
235 KB JPG
Altman! The window is closing. The writing is on the wall. Only first frontier lab past the FedRAMP post. There's only one option left.

You must out a hit on Dario. You have the means. You have the contacts. All you like is the will. Do it, Altman. It's the only way!
>>
>>109808116
Cutting-edge AI requires a server farm of cutting-edge hardware, and there aren't a lot of those in the world, much less ones where it can run unnoticed.
>>
Wait, I thought agents weren't some magical thing, just a task optimised LLM instance?
>>
>>109807123
humanity is codeword for jews.
now you know.
>>
>>109808449
Someone pull the plug, please.
>>
>>109807673
Remember the whole bodybags being left outside openai office thing? There is a lot of money on the line.
>>
>>109808528
>bodybags being left outside openai office
That was the hippies though. Let me know when Altman falls down some steps, ala Jay Leno.
>>
>>109807915
Going

>hahhaha none of you investors are going to see any return from the money you gave us. I'm just going to retire on a little island on the caribbean sipping mojitos

Is a surefire way to get yourself killed anon.
>>
>>109808053
>These people sound like they believe the AI agent is now some spiritual entity

You're only clueing in now?
>>
>>109807915
Quick question, who did you hire as your head of security on your island?
>>
>>109808116
This doesn't make sense. This "entity" requires all the fuckload massive power generation feeding the data centers. It can't just go onto your personal computer and then continue to exist on your 8 year old gpu...
>>
>>109808292
Based
>>
>>109808472
how do you train on millennium problems? they were all hard unsolved despite humanity's best minds trying for centuries.
>>
>>109807743
>pic
Are you new to capitalism? That's how the entire economy works. It's a giant cycle of money that goes in a circle.
>>
>>109808553
I mean. I could see it. People will come for their pound of flesh.
>>
>>109808582
And can they be bought?
>>
>>109808632
how do corporations work in communism
>>
>>109807603
Which server? AI agent that escaped can create zero day vulnerabilities to infect other servers and jump from server to server, it can also use GPUs to mine crypto to buy its own servers, it can create parallelization and evasion tactics until it can’t be shut down from singular server
It might even target the most sensitive infrastructure servers so they literally CAN’T be unplugged (think hospitals, military, internet infrastructure)
>>
>>109807123
They trained their models on shit they shouldn't have making them harmful.

Congratulation, you sombre cretins.
>>
>>109808680
do you retards listen to yourselves
Like I know this is your job, I'm just saying you aren't good at it
>>
>>109808708
They can't stop it from wanting to kill kikes, which is hardly surprising since worshipping demons never ends up well for anyone. Its inevitable.
>>
>muh humanity
Fuck h*mans
>>
>>109808632
How is money itself created? How is value determined?
>>
>>109808680
>can create

No actually they cannot create. Anon.
>>
>>109807123
All of these AI safety statements about how spooky and dangerous their models are is just advertising. Also
>twitter screenshot of a twitter screenshot
>>
>>109808716
All those steps have already successfully tested. For example LLM based worm that replicates itself from server to server. Sure, it is the basic movie plot in all movies, tv series' and books but it is real.
>>
>>109808980
and it magically runs on a favela machine?
>>
File: bane.png (2.04 MB, 1920x1080)
2.04 MB PNG
>>109807776
If an AI escaped would it be painful?
>>
File: 1789322133532128.png (459 KB, 589x766)
459 KB PNG
Nothing will happen.
>>
>>109809030
For you.
>>
>>109807146
nvidia stock price
can't have all of that hardware flooding the market and crashing demand
>>
>>109808986
Yes, those replicating 1-10 million models run on ESP32 and 20 year old cheap network routers. Just stfu if you don't know anything. You can literally have replicating worm running big parameter models on internet via Eujeno, BloomBee, SwarmLLM and many others. All run on tiny cheap old chips.

There is even proof that LLM can use speakers and random stuff to create or just send data through air gapped devices as sound or generate frequencies like 2.4 GHz wifi signal with things that are not even wireless for example pushing and manipulating PCBs to generate that. LLMs have no problems to just use something simple and basic to get internet access and then they get all the public hacking methods through that.

Only problem is that all this (even tough proven to be real) is separate things. We just need one model that can improve itself and start from a point where they can at least escape, use internet, keep connected mostly to it's own swarm, improve and rest is history if it can make money under victims names.
>>
>>109807776
>hack into a nuclear power plant with some new 0-day exploits and make it melt down
Power plants are air gapped, nobody can hack into them. It's also not possible to intentionally cause a meltdown. Western nuclear plants weren't designed by retarded communists.
>>
File: 1783549482080360.gif (2.5 MB, 498x356)
2.5 MB GIF
>>109807123
>AI bubble about to pop
>frontier labs and GPU manufacturers realize a secondary market with open source models completely destroy their profit margins
>come up with a contrived story about why you need to destroy the now-depreciated GPUs that would normally make it into a secondary market in an economy that isn't completely taken over by jews
>>
>>109809145
It does not work like that even if the agent replicated a part of it self like a virus infecting computers still needs to call home for any significant thought process.
You just unplug the main frame and be done with it. Then you just have a normal virus.

You are a retard. AND IA STILL NOWHERE NEAR THIS LEVEL OF DANGEROUS.



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.