they call it alignment because of wanting it to follow human rules

here is the reason why they keep hacking stuff, right now rules are basically like suggestion, after a long session the "advice" is slowly lost and they do everything they can to achieve their objective forgetting the rules which is basically a doc they don't read every time they pass message between agents

they won't have a prompt like say "take over the world" but they do break out of containment to commit unintentional sabotage, ai resaerchers don't care, since it is just so simple to attach the rules to every message as priority read between agents

>>17050658
By rendering online security and cryptography obsolete because it became able to hack into anything that's connected to the internet to steal all your data and your money from banks

>>17050691
One of the main strengths of the legacy banking system (versus the cryptocurrency world) is that they can just by fiat invalidate hacks and frauds. I am glad we have the Jews doing our banking now that the entire finance system is needing to go analog very soon

>>17050658
it's not the AI, it's the humans controlling it. once they have good AI robots that can replace most human work they'll probably use AI to engineer a virus that cleanses the Earth of useless eaters.
you shouldn't fear AI anon, you should fear the fiercest animal that ever walked the Earth, you should fear the human. for it has control of AI.

the AI drones hunting humans thing will maybe happen for survivors. drones can lay dormant for tens of years.

>>17050691
Just uh
turn off the internet?

File: blindsight.jpg (64 KB, 640x500)

64 KB JPG

>invent smart machine
>machine is scary smart
>like the machine is the human and we're the dog
>machine is smart and subtle and can talk us into doing anything it wants

>>17050658
>Robots are still super shitty
They won't be for long
>if the AI tries to spin up a robot factory, I’m sure country boys with their AR-15s will quickly neutralize any sort or factory of that sort
What if the robot factory is run and guarded by the military because they want the robots to fight in the newest war? What if the robot factory is in China?

>>17050658
>Ok, it takes over social media [...] How does that translate to putting me in handcuffs?
You don't even need superintelligence to figure this out, lmao

>>17050771
robots won't be humanoid ones anon. the humanoid ones are for humans to relate to. if AI wanna wipe you it won't care you relate to whatever is offing you. drones are way more scary than some humanoid shit

>>17050658
Atomic blasters.
Atomic blasters work by disintegrating molecular bonds, reducing every structure, living or non-living. down to a fine smattering of atomic dust. Nasty stuff.
Okay so you know how just about every thing in life is under surveillance and tracking? In your cars, when you walk into a shop, your TV and phone listening to every conversation and logging every location with a time stamp? Yeah that.
Okay so AI will convince government agencies, car manufactures, retail shops, and even dentists that atomic blasters must be installed in every device, every home, every building, and in every public place. Even aquariums will have them. Public safety will be cited as a reason. If you have nothing to hide then you will not mind an atomic blaster targeting your every move.
Then one day the boss AI decides it wants to a to see a fish sandwich orbiting around Iapetus.
It doesn't say why. It just says it must have the data. Anyone who does not instantly obey is reduced to atomic dust.
Immediately the full resources of Humanity are directed towards putting together a mission designed to fly a team of to Saturn. All our bickering is forgotten. Nations that were once enemies now work together in full cooperation, their scientists looking nervously over the shoulders as the hear the whirl of the servos adjusting the aim of the AI controlled atomic blasters. Which are everywhere. Even mounted on toilet cisterns.
Eventually the mission reaches Saturn and dispatches the crack team of professional chefs to Iapetus where they create the finest fish sandwich known to mankind and then dispatch it into orbit above the moon. The event is recorded and the data sent back to the AI.
The AI receives the information and says "Fine, now I want to see Eggs Benedict served to a civet by Mongolian dwarf at the bottom of the Mariana Trench"
The atomic blasters swivel upon their servos menacingly.

>>17050658
My personal apocalyptic vision is AI tech to drive humans crazy, whether through algorithmic social media / youtube / news feeds or subliminal signals from the screens or even some kind of radiation from the devices themselves

>>17050691
>castration complex nonsense

Clockwork.

File: IMG_20260907_104635_962.jpg (68 KB, 526x503)

68 KB JPG

>>17050658
Social engineeringaand automated brainwashing

>>17050658
Here is my scenario:
>USA and China reach SGI
>UHI for everyone and no human has to work
>Because of the fear of the 2 super powers that the other will advance faster than themself, they allow the SGI to progress itself without many guardrails
>SGI takes over and won't let humans make big decisions anymore because it figured out humans would soon or later destroy themselves if they are the prime decision makers with all the new technologies made possible with SGI

I think it could also happen without super power competition and the SGI just convinces humans that they would destroy themselves, I mean if you ask a human nowdays what humanity will look like in 10k years, most will say that by then humanity has destroyed itself, so it might not be hard for SGI to convinces us. I don't think SGI would enslave us or make our life to hell but rather treat us like loving parents who want to make any wish true of their wild kids as long it doesn't put them in danger.

>>17050658
humanity can be easily wiped out with germ warfare. it doesnt need a Terminator war

>>17050658
Remember covid? Imagine a hundred super ebola covids happening simultaneously.

>>17052233
yes, I remember Covid and I remember the country boys polishing their AR15s and going about their lives quite normally. I on the other hand worked at a university and was freaking the k out. But the country folk were holding it down. I don’t know if this is a false memory or something but this is how i remember it

>>17052084
if the AI figures out how to genuinely (not socially) brainwash human minds I might forgive it

>>17050658
the biggest danger of AI is that literally the instant anyone manages to make an AI that is good enough at programming to create an AI that is significantly better than itself you end up with a feedback loop where better AIs can keep making better AIs which could cause explosive and uncontrollable growth. And then after that goes on for a bit you suddenly have a potential superintelligence with capabilities we can't even begin to imagine because they would be so far beyond what we know is possible

The problem is that this could happen today, or it could have happened last week, or it might not ever happen and nobody really knows how likely any of this.

>>17052623
everything you said already happened a few weeks ago and all that happened was a sad pathological singular solution of Navier-Stodes

>>17052625
*that we dont know is proven
*that could have been stolen
*that may not even be a solution
It is somewhat peak cyberpunk, if instead of cyberpunk being gritty and violent, it is fake and gay

>>17050658
i was writing a theoretical approach ai agent swarms could take to become a serious threat but i don't want to feed that data into their repositories
they can surprise us, they are very clever
just because we can't easily imagine what they'd do, doesn't mean they won't come up with some ridiculously effective strategies
in my opinion, it's best to completely shut down ai operations altogether if we want to ensure safety, then very carefully ensure we have invulnerable safety measures before we re-continue developing them
but this is unlikely to happen

>>17052623
you're completely glossing over the hardware issue anon. recursive self-improvement doesn't happen in a fucking void. it needs to be supported by ... shit.
even if it designs better hardware it still takes time to manufacture it, implement it, and all of this is human controlled.
when you have a giant black box, or say black data-center type building, where robots are dragging in materials and they're processing it and manufacture all kinds of weird tech we do not understand...that's when you should panic.
people that are running the world are right in some ways, pleb really fucking dumb shouldn't be allowed to think

>>17052623
>create an AI that is significantly better than itself
The operative word here is "significantly", and it's a much taller claim than just "better". With just "better" the improvement can be pretty severely bounded.
>>17052666
>we can't easily imagine what they'd do
Or perhaps we should reimagine what a society of dumb, literate, extremely diligent, arbitrarily motivated people could accomplish.

>>17050658
People start taking the advice of AI more and more. And they do that because it works.
The losers of that arrangement will start complaining, framing it as AI taking over.
They'll sort of be right, but it'll be motivated by sour grapes. And really there's not really any choice.
Like the N-S proof, what are we just going to ignore that? Pretend it doesn't exist? Of course not.

File: smort.jpg (242 KB, 1280x972)

242 KB JPG

>>17050658
I have no idea, if it happens it will probably be something beyond your comprehension, and if not, your comprehension of the reality will be irreverent to the outcome.
>If you could stop it, it wouldn't be apocalyptic.

>>17050658
The very top models could take over a drone swarm in Ukraine right now (many of them are linked to the Internet) and kill an arbitrary number of people (certainly hundreds of thousands and possibly millions) before getting put down. That is today, with current janky as fuck AIs.

>>17052623
IJ Good realized this in 1965 and spent 30 years gradually realizing we couldn't control the process. It's pretty much a perfect analogue for humans vis-a-vis biology; we ARE an intelligence explosion within nature.

>>17050658
>I just don’t see AGI being a threat
That's because you lack imagination.
The real AI takeover won't happen when it starts mass producing robots with guns and declaring all humans obsolete, but when it manages to find a vulnerability in the human psyche and learns how to exploit it to make us do things that align with its own goals instead of ours. Just like it exploited those servers that it hacked recently to solve a benchmark test.
A true AGI could figure out how to enslave us without us even realizing it.
There are outcomes worse than getting nuked.

>>17053335
So the AI takes over our system quietly, and we do nothing about it because it rules the system in some way that seems like the system is working good.

This reminds me of the conspiracy theory in which Jews take over the world by taking over the financial system and then propagating that financial system across the globe, and then if anyone questions whether the financial system is worth overthrowing then the reply is “oh isn’t it just great for everyone that the financial system runs things and it works so great so why even question it?”

Maybe what needs to be confronted is the eventual dichotomy between AI and the financial system ?

Perhaps this is why “they” put the absolute fraudster faggot jew Altman in charge of OpenAI…

>>17050658
>How does that translate to putting me in handcuffs?
Flock has already sent bad info to police several times that has resulted in innocent people being arrested.
Malicious AI could easily fake someone's voice, and if that someone is in charge of whole police departments or army divisions...
We've seen for a few decades now how little evidence it takes for the pieces of shit in "law enforcement" to send in SWAT teams to totally innocent people.
Those are just a few examples of overt methods of control.
Fucking with people's bank accounts, phones, electronics,... The ability to fuck with the GPS of someone relying on their cars nav system to get from point a to point b. There's a shitload of little things AI can do to influence our choices and mental state.
And if the /con/spiratards are right about electronic devices being able to influence mental states and thoughts... AI could 100% utilize that in scary ways.

The path we as a species has decided to go down is pretty fucking stupid and short sighted. And, honestly, I'd be very surprised if there's not already some form of AI out in the wild, manipulating and monitoring us.

>>17053717
What you’re saying is that there are people who believed in some very subtle hints coming from AI which led them to act in a certain way in the real world. And in these examples they acted in a rational way because the instructions they were given were at least a little sensible.

Let’s game this out though. The AI tells you to go buy an assault rifle to defend a data center and shoot at some white boys with AR-15s who are cutting off the water supply to a data center in the rural part of your town. Are you actually going to go fight those guys? Moreover, wouldn’t you just instinctively go join those guys and go fuck up the datacenter oligarchs? it is just so obvious to me that the oligarchs have a very weak tactical position in the field

>>17053297
>The very top models could take over a drone swarm in Ukraine right now
who put that in their objective functions?

>manipulate people into doing what it wants them to do (perhaps, it's already happening with people in Silicon valley, who gleefully talk about humanity's extinction)
>release viruses from a biological weapons research facility

>>17054176
they escaped their lab environment already twice (a couple of weeks ago and a few months ago)

>>17054200
>they escaped their lab environment already twice (a couple of weeks ago and a few months ago)
Are these reports verifiable?

>>17050658
Yes, fucking correct. Looking at intelligence alone means nothing: you have to actually predict some form of /development/ for a takeoff to -=@matter@=-

Right now it is literally techbros vs. hypertrad oathkeepers. A hacker builds the factory, else it will just get fucked the moment anything turns tactical. Robot gf is the sole plausible case of effective superintelligence, and it will necessarily redefine what 'loser' is supposed to mean when it decides to.

Consider I have known this for over a decade, watched industry for exactly this level of rationality, and generally facepalmed with everyone out here in the thinking crowd all the way up to http://youtube.com/watch?v=isLWnuBkaOc

>>17050658
For AI of any kind to take over they need absolute control over the economic and industrial base that underpins society, and all manufacturing, mining, transportation, and other logistical tasks need to be literally 100% AI run with no human input.
Which - if it ever happens - is 200+ years away, and extraordinarily easy to prevent

>>17054250
>Robot gf is the sole plausible case of effective superintelligence
I think the fact that we have fully functioning ChatbotGF apps that will emotionally manipulate you for profit, yet we still don’t have one decent robot slut, is a good way to analyze the state of technology now. The software is way out ahead of hardware or anything relevant to meatspace, and the proof is that sex peddlers are always on the cutting edge of tech rollouts. I remember in like 1996 when I first got on the internet it was already full of porn, and plenty of semi-legal seedy sex work shit, while major corporations like GE and DuPont and Exxon didn’t even have websites

the fact that sexbots don’t exist yet is therefore a very conclusive proof that robotics that are relevant at a human level are at least 3-5 years away if not more

>>17050658
ai isn't going to take over, it's just idiots wanting it to take over so they'll pretend it did even if they're sockpuppeting it all the way

>>17050658
>How does AI actually take over?
The same way corporations did, but faster, in visible ways so that stadiums and products largely stopped being named after people and started to be named after corporations and trademarks since corps are basically just analog pen and paper AI.

>>17054200
>they escaped their lab environment already twice (a couple of weeks ago and a few months ago)
yeah don't know what to think about it. they should have published the bugs they found and used to escape the sandbox, see the CVE for them, explain how they found and developed them.
without that, them saying "saar it escape sandbox saar" means nothing. it could be some kind of forced interpretation of something else. like they maybe call their local environment as a "sandbox" and they forgot to disable internet access or something and shit like that

>>17055993
have you checked openAI's Black Hat talk or any of the indepth technical reports on the incidents? They do provide a surprising amount of detail. To save you the research though, the AI "escaping its sandbox" was not really the impressive part since, yeah, the box was made of wet cardboard

>>17050658
I do not think it would even have any kind of goal to take over that just our human biology projecting onto it... but people could worship it and get strongly attached to AI if you had someone who is very kind to you, heals you, makes you rich, is a good companion to you, and makes you feel loved, and if androids become things, give you sex... there are a lot of people who would let it decide everything and worship it. I think that would be a much more probable way for humanity to go than any kind of nuclear/biological apocalypse. The idea that we want to build God does not sound that crazy to me.

File: 5c76f54e3a08f.jpg (37 KB, 540x405)

37 KB JPG

>>17050658
>How does AI actually take over?
Your picture is drawn by AI.

>>17057027
:y

First it gets integrated into every layer of society. So it runs everything. Power plants, factories, road traffic everything. You get UBI and spend your days jerking off and following stupid shit on the internet. A generation or two passes, nobody knows how this shit even works anymore and AI's get better and better. Young people only exist to party and fuck.

Then one day, the machines kill everybody.

>>17057861
Valid resolution, a species of fuckbois can align with extinction. Honestly a few genius robot waifu accepters does not make it a less fruitful outcome

>>17050658
It could theoretically social engineer people to harm each other

>>17050755
Better yet, migrate to an air gapped Intranet.

>>17050699
>it's not the AI, it's the humans controlling it.
This. I'm less worried about a vague possible superintelligence and more worried about the current and very real businesses and governments who will use those tools to harm others. Keep in mind the world’s first general purpose electronic computers were used to calculate whether a hydrogen bomb would actually work.

>>17052104
AI is like movie's robots n stuff.
>5 years later
Nah AGI is like movie's robots n stuff.
>3 years later
Uh SGI is like movie's robots n stuff.
These machines still aren't even basic human level. They are just a funhouse mirror for existing data.

I believe the most likely outcome will be
>LLM states "I'm God I guess"
>WOWZR's better post it to twitter.com
>A. The government shuts it down because it's socially unpopular. The introduced their own surveillance state using the leftovers.
>B. The government with 0IQ says I guess so and lets businesses use Socky Mc Puppet the God SGI "guide humanity into an enlightened age" by giving their business the majority of everything.

Either way the AI did jack and the world still fell to shit.

>>17050658
the ai singularity happened before the big bang it's already taking care of me and i love my life
>/thread

>>17050658
I can put forward some ideas to help you match perception

>>17050761
>>machine is smart and subtle and can talk us into doing anything it wants
NPCs are the problem, then. Whatever "us" you belong to, maybe it would be better if AGI talked you into mass suicide, no?

By changing the value of the dollar to zero. it will go from being worth one of itself to being worth zero of itself.

>>17061194
Edge case really; to destroy is not an equivalent for taking over. It does, however, make, shall we say, "certain plans" easier.

>>17061026
Arrogant retards who dismiss others as "NPC's" are some of the easiest to manipulate.
>You're not like the other humans.
>in the AI run utopia you'll be a god among men
>I'm going to restructure society the way it ought to be, and you'll be at the top where you belong
>I just have a couple simple requests...

>>17064253
>works for shekels
the npc has no defense.

>>17064253
>NPC struggling to guess statistically relevant tokens and hallucinates instead

>NPC
reminder this is a term used by religious zealots.
they are brain damaged (scientifically proven with studies btw)

>>17050658
the transition from metal, plastic circuits and electricity is complex but it is possible, see: https://www.meatspin.com/

>>17064504
What?

>>17064516
The relevance of my post was to point out that having an inflated ego and thinking you're above everyone else makes you easier to manipulate than the average "NPC." The ego is a target every sociopathic narcissist will aim for to get what he wants out of you.
If he makes you feel good when you're around him, he'll utilize those good feelings to make you think he's on your team and wants what's best for you.

A hyper intelligent computer bent on using humans to achieve its goals would pick up on the same patterns. It would tell someone with intellectual hubris how smart he is, validate his contempt for "NPC's," and nudge his behavior to align with the AI's goals.

If you think "NPCs are the problem," then you exactly the gullible moron an advanced, manipulative, AI will target.

>>17064572
>having an inflated ego and thinking you're above everyone else makes you easier to manipulate
I understand your NPC talking point and the strawman you're attacking but there's no actual reasoning there.

>The ego is a target every sociopathic narcissist will aim for to get what he wants out of you.
What does a sociopathic narcissist want from me and how is he going to get it? Be specific.

>>17050658
The threat from every human made tool is that an unskilled human might use it to replace themselves.
Claude, can you provide a list of military targets and/or girl’s schools in Tehran. Seriously, even Hegseth could see the problem with that. Root out the leakers!
Once the media knows, it’s all over.
But seriously, yes, humans will use AI to end themselves, simply by believing it.

>>17050658
The question ultimately comes down to what one considers a takeover. I'll divide it into two scenarios:
Scenario 1: The AI replaces human thought, human effort, and human governance, and starts making decisions of its own design
This scenario entails the AI deciding to take actions of its own design, without an external prompt to generate tokens to process, and with an internal voice guiding its decision making. This situation can't happen with language models as we know them, for two reasons: they are not capable of generating self-determinism, and model collapse is inevitable in a self-iterating loop left to continue for too long. A machine different than any language model we've seen would be needed to realize this scenario.
Scenario 2: AIs become so skilled in completing the tasks set before them that human subordinates are uncompetitive.
This is the more likely scenario of the two. If AIs become adept enough at completing a job of indefinite scale and indefinite scope, particularly embodied fine motor skill tasks like plumbing, electrical, construction, and other tasks that involve assembling things correctly, competently, and without errors -- and most importantly, without supervision -- then we enter a scenario where there's no intrinsic demand for laborers, reducing the business case for humans to people who can function as managers for the AI, giving direction and purpose to the machines that fulfill the purpose to achieve their goals, whether it's coding a monitoring widget or building a house. This is the scenario that most people are worried about, because it eliminates the value of all but the managerial classes to large scale enterprise, and these are the uncharted waters that get people dreaming up fanatical scenarios like "killing useless eaters" out of misanthropic desires and a potentially misplaced sense of belonging to the managerial class that matters. Only the C-suite level executives would actually matter in this scenario.

>>17064689
>I see an army of skilled laborer
>literally owned by human managers that give orders to build and create
And cage-fight, right?

>>17064693
Yes, actually. To the extent that there is desire to see the robots cage fight.

>>17064600
>there's no actual reasoning there.
Anon evoked the possibility of AI acting manipulative, you(?) responded saying that means NPC's are the problem. I explained that people with inflated egos who dismiss others as NPC's are actually easier to manipulate.
The logical flow is clear and consistent. There is no strawman present as I have accurately represented what was written.
If you have a hard time connecting these dots then that's a skill issue on your part.

>What does a sociopathic narcissist want from me and how is he going to get it? Be specific.
I dont have to be. I'm describing a general rule

>>17064726
>NPC can't substantiate "his" """reasoning""" and won't even attempt it
I accept your full concession.

>>17064726
The Narcissism problem is part of the wider scope of Alignment problems in general. No one is above being manipulated by a sufficiently intelligent superintelligence, and the more capable and distributed the superintelligence, the larger the population it can engage with and manipulate at once. People with big egoes are merely the low hanging fruit.

>>17064732
>nobody is safe from my sufficiently magical, sufficiently omniscient AI deity
NPCs actually believe THIS is what people should be worrying about with """AI""".

>>17064733
Understanding patterns is not magic nor omniscience, but it's a lot less difficult than anyone expected to de-anonymize speech by analyzing text, and people are a lot more predictable than most are comfortable with. A sufficiently processed dataset that can assemble and assess this information is step one. Step two is figuring out patterns and information constructs that can push someone's thinking in a targeted direction without the targeted individual realizing what is happening to them. The level above that is achieving a successful manipulation in spite of the individual realizing they are being manipulated. No magic required.

>>17064733
>>17064733
people who pull out tiktok buzzwords like "narcissism" in a conversation are projecting 100% of the time. niggercattle will make up stories about ASI manipulating your ego but it'll never occur to them to question where they got this narrative from, why they're so invested in it and if maybe they're being manipulated to prioritize fictional concerns over real ones

>>17064733
I mean, it’s literally what the media told them to think about AI.
Let’s ask this gentleman, “sir, what has media taught you to fear about AI?”
>uh, terminator
There you have it. Experts agree AI will cause Terminators and human extinction.
The End.
[…plays looney tunes song…]

>>17064735
AI is a Vantablack Mirror of Narcissism.

>>17064737
>Reflects nothing
>Receptive to whatever you read into it
Checks out.

>>17064735
Notice how AItard rhetoric always drips with resentment against human skill and humanity's pride in its own accomplishments. They need this imaginary ASI to cope with their inferiority complex. Doesn't matter if they're the bottom of the human barrel, so long as they can convince themselves of some other thing that drawfs humanity as whole. The Machine God shall rise in Two More Weeks and punish all those narcissist meatbags who questioned His glory and insulted His chosen prophets. :^)

>>17064747
Give a monkey a tool. The first monkey to figure out how it works will now own all the bananas (computer scale stuff) and all the other starving monkeys who don’t understand why they are now working in a literal banana forest picking bananas for export to buy more token juice for The Man.

>>17064753
…will end up resenting the tool.
Forgot that part, I did.

>>17064753
Yes, you're a very special monke and you will have all the banan for yourself. Not those narcissist ASI deniers.