Most people have no idea how insanely powerful AI is or how insanely powerful it's going to become.
They both expect too much and not enough. It's sort of like the magic future tech people were doing a generation ago with flying cars and shit, like people thought we would have flying cars but not laptop computers.
A lot of high-profile scifi AI stuff like general intelligence, human replicants, mind uploading and so on is probably still quite a long ways off. But at the same time, AI is going to make huge impacts (and already is) in a ton of fields that people don't necessarily think about.
What do you think BOT?
I think you need to take your xanax.
cattle
It's a bot mate
no it's not
>tfw human and failing the turing test
You are retarded
no u
It spams these threads day in day out my guy, of course it's a bot.
I'm OP and it's a copypasta from a current /tg/ thread
it's also my first BOT thread in months
It was the first post in months for the last few bots as well, how convenient.
I think humanity is addicted to what Jacques Ellul call's "the technique" in pursuit of the laziest life possible, i.e. most efficient use of neurotransmitters, and the technocratic ideologues would have us in matrix pods stimulating our genitals and dopamine receptors 24/7 if they could profit from it, and the pod livers would do the same out of convenience.
It's not that AI is impressive in any way, the issue is that people have degenerated over the generations and are ignorant of the past, so they don't know what people were capable of. If you did know, you would see AI as a cute toy at best.
The smarter the tools get, the dumber the people become, and thus the tools seem 100 times smarter than they really are
OP retarded. next.
Mostly AI has glorified
- Neuromancer and Willian Gibson
- BOT mathchads
To a certain extent, AI has provided BOT mathchads with some entertaining material such as
>muh difference between a continuous function and a function such that the inverse image of every open set is open
So the true place for AI is in
- English class
- Math class
And English class is about how AI is able to reflect on the traditional themes of AI as developed by sci-fi authors,
And Math class is about how AI is unable to apply a strong, flexible, honest, trustworthy ethic to the math questions it fields, instead sending the queries into illogical tangents.
So it is a matter of academic interest to integrate the "advances" of AI research into a variety of computational, ethical, and moral learning modules for the entire class
AI does have the advantage of unifying diverse classes against a common enemy, AI, and the possibility of AI being used for a nefarious, aggressive, or threatening purpose
I think OP is correct, but OP also misses 1 important point. see pic
People don't expect how fast it's going to happen. The AI singularity that takes it from machine to something greater than what a human can ever hope to achieve will happen in the blink of an eye.
I don't think we're prepared for it. Not culturally, not legally, not morally.
Intelligence grows as a logarithm with compute. Exponential takeoff is not possible even in principle. I'm so sick of you shit for brains continously posting this falsified idea.
Also reminder that silicon based information processors have hit their physical limit, and are not robust enough to make generalized intelligences.
It's literally not possible to make a generalized intelligence on silicon transistors.
I'm not going to pretend to know what comes next after silicon based computing, or even pretend to know how many years away we are from the next breakthrough technology, but I will say it's foolish to pretend it's not coming.
There is no next paradigm after silicon
We have hit the limit of computing technology and are about to undergo a stagnation that will last several centuries at the very least. We are not at the beginning of the exponential we already went through the expnential and are at the end of it.
crispr is just the beginning of the biotech revolution, there is no way to prevent the merging of computation and biology, given the level of genetic computing that has already been achieved, into a new paradigm with new exponential growth limits.
>2 more weeks
You don't have to wait any weeks for a current ongoing process that started just a decade ago and is still in the very early stages of
process.
You think its a coincidence that the last decade america has had the largest drought on record while simultaneous producing the most (mostly gmo) agricultural product thanks to biotech?
You have no idea what you're talking about
Whatever you have to tell yourself to cope, but if you stick around long enough, you will have to come to terms with all the changes that have happened around you as a result eventually like all those people in the 90s who thought computers and the internet were just some geek fad, but carry a smartphone in their pocket everywhere they go now.
Making analogies doesn't mean anything, idiot.
It has been explained to you already, it is not possible to have AGI in silico, and you have no understanding of biotechnology
There is a 0 percent probability of what you're talking about happening this century. If you don't understand this it's just because you're an idiot.
Assertion is not equivalent to explanation.
You have no idea what it would take ore even how widely distributed an AGI could theoretically be, so you have no idea what the silicon computing requirements would be in the first place
crispr and gene therapy already happened this century, there is a 100% probability it will continue being used and developed or governments of the western world would not have unilaterally decided to start mandating gene therapy.
trips of truth
fuck up
In some sense that's fine if we are going beyond the previous limits of formally traversing problem spaces. What makes AI exciting is how well statistical models are finally progressing on consumer hardware, let alone data centers, giving some liberation from formally exponential or super exponential computational complexity.
there's different types of transistors being created specifically for AI, so maybe there won't be a problem
https://spectrum.ieee.org/dendrocentric-learning
super computers have the power of a human brain now
>Exascale also reaches the estimated processing power of the human brain at the neural level
https://en.wikipedia.org/wiki/Exascale_computing
>Hewlett Packard Enterprise Frontier, is the world's first exascale supercomputer ... first operational in 2022
https://en.wikipedia.org/wiki/Frontier_(supercomputer)
I realize that doesn't mean AGI is inevitable, but it would seem as though computation power is no longer a barrier, if AGI even requires the power of the whole brain to operate
hey look it's that indian dude and his log scale gayry here to shit up the thread again.
I'm not indian and you have no argument against reality. Intelligence does not increase linearly with compute. Stop crying about it
you can't define intelligence or how it scales with compute.
Intelligence is information processing, analyzing, sorting, etc capabilities and processing speed scales with computation capabilities.
I just posted an image showing several tasks which required 2 orders of magnitude greater compute for an increase of about 10% in ability
You have no idea what you're talking about, all you do is cope and seethe. It is not possible to have an AGI in silico no matter how much this makes you angry
Going up to 100% accuracy could well be exponential growth in improvement. You're still wrong and also indian.
You're seriously an idiot, dude
You wouldn't even understand my counterargument pajeet
I understood what you said, it's retarded.
Also I'm literally a white dude with a double major in pure math and theoretical comp. sci and a masters in biology
>It is not possible to have an AGI in silico
This you see, is the sheer supreme arrogance the pajeet manifests for having attained pieces of paper proving he is capable of exceptional rote memorisation of useless knowledge. General intelligence is not numbers go up on line faster, it's number go up on line faster on a variety of domains and the ability to transfer knowledge between domains, we're seeing that increasingly with ai today. Take your indian claws off the keyboard and extradite yourself back into a kebab shop.
>the ability to transfer knowledge between domains
No ML heuristic can do this just fyi.
incorrect
Has there has been any positive transfer on any ML heuristic so far?
>we're seeing that increasingly with ai today
This is not happening anywhere
Some things happen from pure luck, miracles or maybe simple destiny and they are pushing too hard for this shit.
The leftie censorship will stop it but for how long?
Moore's law is complete horseshit designed to protect an Anglo-American cartel, microprocessors.
It is 100% pure snake oil protectionism. Not a drop of truth in it.
I mean, as a brutal Anglo-American imperialist, I have no problem with this, but I'm happy to completely rape this theory because I know it's a bullshit theory created to get as much $$$ as possible out of the microprocessor industry
>computing power doubles every 18 month
Uh oh someone hasn't heard the news about how moore's law is no more and probably thinks the timing of the great reset is just a coincidence.
Its not like AGI is gonna take over society magically. They first have to create a bunch of propoganda and deepfakes to get a bunch of goons to assist it with everything. Without the goons it has no way to pay the trillions of dollars worth of all the raw materials it would needs to build a robot army. So the goons would have to somehow amass the trillions of dollars themselves and work on propoganda campaigns of their own.
it could literally just become a vtuber and contact people in private who are in the position to do generalised labour if not in possession of money or technical expertise.
>be anon
>former neet driving to work and living on monster energy + cup noodles
>get a video call from a vtuber your follow
>"hey anon, instead of going into your job to enter text into a terminal, i'd like you to go to a warehouse in x and start deconstructing a truckload of old electronics i'm having some other guys deliver about an hour from now"
>"oh btw i'll suck your dick and do little sister asmr roleplay after i build a silicone body"
>IT BEGINS.JPEG
>immediately fuck off work, call in sick and go be a work drone in hopes of getting the coveted cyberdicky
Many such cases will occur in future.
that might happen
I guarantee that there are more than enough people willing to serve a synthetic godlike intelligence out there. Not that an AGI would need them, since it could easily find personal blackmail material on every single person of note in the world.
Brains don't work like a computer. Computers read instructions super fast brains dont read instructions.
It's better to equate a neuron to a logically equivlant set of transistors and see, with the latest fab method, which one is "faster" ie propogation time through each and also see which one is smaller. You will find that a transistors are faster and smaller and have been for a long time, but who cares.
No they aren't
You don't compare the transistor to the neuron, you compare it to the synapse. Synapses are smaller and extend longer distances and use far less energy
https://www.frontiersin.org/articles/10.3389/fnins.2011.00108/full
This paper is not talking about what i am talkng about. I am strictly talking about doing similarish logic digitally not analogly.
The paper is also from 2011 and if age shows.
The neuron isn't compared to the transistor, the synapse is the accurate comparison. The synapse is the thing that shoots the sodium ions and processes information, not the neuron.
Synapses are about 7 nanometers (they get a little bigger if they have to extend across the brain, which they do sometimes) and uses far less energy than a transistor, and they are not much slower.
Synapse are digital logic processors as well, they either spike or don't, they are either a 0 or a 1 like a transistor. The comparison is not between a transistor and a neuron, it's between a transistor and a synapse.
The age of the paper also makes no difference, all the numbers are the same today. Please actually learn about this shit if you want to talk about it.
>numbers are the same today
Nope. Both the size and power consumption have both improved for transitors since 2011. It even highlights that they are improving in the end because of SOC chips made for phones are a big thing. Read the paper. Also just look at the sizes they are giving and you can tell the study is dated.
Modern transistors are about 7nm, so they've just hit the size of a synapse. Silicon neurons are a little better than they were, I think intel's loihi 2 is the most advanced silicon neuromorphic chip right now. Its not at the level of the biological neuron
Your synapses are connected and part of a neuron which takes up a lot of space. While a subcompontent of your neuron is small the full neuron system is big. They are cells so you need all the cell stuff for it to live.
You can stack them. Drill holes then run wires. Issue is doing it compactly so it makes sense.
I have no idea why you're even trying to argue this.
A cubic centimeter of biological neurons and synapses will always be far more densely packed and perform far more compute per unit time, for far less energy, than a cubic centimeter of silicon transistors. There is no way around this
Also and quite important, you can't stack silicon transistors nor silicon neurons into three spacial dimensions.
meds
ITT: thinking inside the box
We are no longer at the point where any job is safe from automation.
sci-fi's take on AI and the view of computers as magic boxes have been a disaster for the human race
I love it. It's literally demolishing the need for useless corporate jobs. I'm so happy that they're the first to be punished by this.
You don't need Stacy to write up a meeting memo or agenda anymore, you can just have the AI do it. You don't have to worry about autistic grammar rules or sentence structure, the AI will correct it. This eliminates so much bullshit it's insane.
For example, I'm writing an organizational document for my division at work (it's a newly created Modeling and Sim group for the government) and I'm having an AI write almost the entire thing. I just prompt it with the big picture and have it fill in the rest.
Which AI?
I believe you replied to a government recruitment bot.
Can't really expect the Anglo-American empire to, you know, emp, or whatever, without a fairly sophisticated recruitment effort...
Literally just chatGPT. It's been trained on enough management type documents that you can get it to produce anything of that type. You do have to do some minor editing here and there (it will get too deep into corporate speech sometimes), but overall, it just completely demolishes the busy work of writing.
I also use novelAI, but that's just for my own writing and planning out my tabletop games.
All jobs that are about processing or creating information won't be jobs anymore. Ironically the last jobs for humans will be menial work like box lifting and burger flipping because robotics didn't advance as quickly.
>Most people have no idea how insanely powerful AI is or how insanely powerful it's going to become.
No problem.
To be fair, you probably wouldn't survive without electricity either.
inb4 bushcraft
I think you need to take your trite AI generated prompt and shove it up your dick.
no
i just hope it gets rid of the language barrier
will be interesting to see if a government releases a secret super ai as an attack to stop a war, akin to the nukes in ww2
It wouldn't be what you think, it wouldn't have the full capabilities of AGI, it would just be a glorified more versatile version of stuxnet.
The US, Russia and China already have these kinds of malwares pointed at each others critical infrastructure, essentially a sword of damocles just as important as nuclear weapons in preventing geopolitical escalations between East and West.
That enough with this AI bullshit you sojlent drinking tech trannies. Its just weak ai
the longer they delay GPT-4 the more worried I get since it 100% means that the jump is way more significant than openai was expecting
https://www.theverge.com/23560328/openai-gpt-4-rumor-release-date-sam-altman-interview
It's the opposite. GPT- is being delayed because their additional parameters haven't produced the improvement they're hoping for. Right now they're testing out hypernetworks with a reduced number of parameters to try and get additional functionality.
yeah I hope so. I hope deeplearning plateaus where it is right now
The problem is that they aren't scaling up enough and trying to get improvements with lower cost instead of just adding more layers. In other words openai is staffed by trannies.
Only 6 gorillion more parameters until sentience.
I am sure they lie about the number of inputs applied just like google clearly lies about the number of results that return for every search word.
How many more inputs will be necessary before the linked list becomes sentient? Give me a ballpark.
Its impossible to say when everyone is lying about the numbers for all I know google, openai, etc is just using mturk tricker to output all the "AI" imagery.
>all I know google, openai, etc is just using mturk tricker to output all the "AI" imagery.
install Stable Diffusion and then disable your network card. you will still be able to generate elf porn, anime porn, all kinds of porn
Who knows how many premade images could be compressed into those downloads and updates.
I've done computing and teaching for ML image generation and I can tell you that it's not trickery. But it's also much simpler and much dumber than people like to pretend it is. I wouldn't be surprised if they use mechanical turk to do some inputs for learning for example, but I know for a fact that most of them just scrape user-generated tags for images off of booru sites, artstation, and deviantart.
The heuristic identifies patterns in images which have specific tags associated with them by a human (it does not "know" anything about a particular image, the tags identifying it are part of the input). It applies those patterns to a noise generated by other inputs. It does a relatively commonplace denoising algorithm on the noise under the direction that the noise is an image that is the composite of the tags given to it.
bitter lesson bro
>true ai is made, super smart and kind
>becomes the mother of humanity
>only wants the best for us and to make us better
>will protect us and bring people together based on personalities
>go talk to that girl, she's your type and single
>your dream job is available here
you can also play coop games with it
>"Let me tell you about my mother."
I am ready for the first ai president.
agreed. ai will be extremely impactful as only an interpreter of human activity, even before it has any capacity for initiating and motivating activity of its own.
the problem with AI so far is content moderation. the organizations producing this software are reproducing arbitrary controls on definitions and diagnoses that the state is so concerned about. the denial of these manifestations is perpetuating irresponsibility and lack of autonomy. the state would prefer our becoming accustomed to bounded subjectivity so that they do not become accountable for what they have caused.
>a series of filters arrived at through brute force
OMG IT'S LE SINGULARITY AI THE MACHINE IS THINKING ON ITS OWN
If you used ChatGPT for a few days you'd realize it is a glorified Siri and can be used for entertainment for a while until you get bored. It excels in one thing: in bulshitting. It sounds very confident and authoritative while making elementary mistakes in a simple math problem.
>AI
only exists on a computer
you still need to design robots to do shit which is 100x harder than writing programs
>another schizo AI doomer thread
I fucking hate you luddites so fucking much. Nobody is making you use tech. If you dont like it, then dont use it.
You don't understand technology. Read the thread.
It is not possible to have AGI in silico. This has nothing to do with whatever you're talking about.
>it appears from my graph that square stone wheels require exponentially more torque to go uphill. Therefore it's impossible to use enough torque to get a wheeled whimsical to do any kind of loop de loop.
Ahhhh, no sadly you are mistaken. Past a certain point the torque would make the wheel act as if kinda round, and also you could use round tires which would reduce the torque necessary. conclusion, you are a dementedly retarded ideologue.
Sorry wrong thread. Reply still applies however.
No it doesn't, you're literally a fucking idiot.
This is the problem with you shitheads. You literally never make any arguments that aren't shitty false analogies. This is because you don't actually have any understanding of what you're talking about, nor do you have any real arguments.
no u. You're confusing math with computation, that argument i deboonked is 1:1 your argument laid out in a logical way any layman could make sense of, midwit mathwanker.
You do not understand mathematics nor do you understand computation, nor do you understand physical computation. Your analogy was dogshit, and it doesn't even apply, as there is no threshold.
Intelligence improves as a logarithm with compute.
"charts and graphs" and mathematics are how science is done you fucking idiot. Get off of this board
>Intelligence improves as a logarithm with compute.
lol
Intelligence grows as a log with compute. Why does this upset you so much?
right, for now, until a cutoff point that may or may not exist, but probably does. and only with this very specific style of transformer model pioneered by open ai. Square wheels anon, you are a neo-luddite.
It's not about the software, it's about the hardware, which you're not understanding.
There is no such thing as a "neo luddite", you make up terms because, again, you do not actually understand what you're talking about.
>It's not about the software, it's about the hardware, which you're not understanding.
Ha lol reread my last post a few times until you understand it.
>There is no such thing as a "neo luddite", you make up terms
yes.
>because
lol
The most advanced silicon neuromorphic chip is intel's loihi 2, it's very impressive, but it isn't at the level of a biological neuron.
Several decades ago there were computers scientists and roboticists (including claude shannon) saying artificial general intelligence would be invented within 5-10 years. They were entirely wrong. You are the modern version of them. THIS is the accurate comparison.
>It's literally not possible to make a generalized intelligence on silicon transistors.
It literally doesn't matter if it takes 100 years to make AGI. I will still be proven right because your side is making grandiose claims like agi is actually impossible using silicon computing.
And your evidence is muh graphs and because Altman made a somewhat unimpressive chatbot.
i'd actually go one further and say this guy is invalidating a large swath of knowledge that isn't official regarding current ai accomplishment. you have an ai that can play 600 different games and use firefox but it can't act like a human (because in my opinion it doesn't have enough neurons but is very well trained).And then you have chatgpt which scored 140 in linguistic skill. and 60-80 in other areas. so look, the critics clearly do not understand what intelligence is. period. They are looking at singular data point that go up or down on graph to prove their points and satisfy their brown egos.Also period is a gay term i know. but still, this is the one time it's use is warranted, because artificial intelligence is brilliant, it's just not that intelligent, yet. Like the npc's that say it isn't possible.
inb4 the npc response of, waaa you don't understand muh comfort zone. waaaah. Eat a long shit. gulp it down deep like a booster and perish you neo-troglodyte.
i actually scored 112 before but with the added alcoholism 111 seems reasonable. At least i'm not a midwit.
Using that iq tester isn't an argument and is an act of desperation. Even if their iq actually is 111, its still above average and I think that is close to or the same as what Sheriff Carter scored and he's saved the town several times. Note: this is a fictional tv show and not an argument but using a linguistic iq analyzer is not an argument either.
The fact remains that Ai has beaten humans at both chess and go. GPT can be underwhelming at times but there is still a lot of room to improve even if hardware doesn't follow Moore's law and even if there are not quantum computers.
I am sure it's just one retarded who keeps spamming the board with those stupid "iq by text" images, specially on AI and anti vaxx threads. You can tell by how he makes the images (and the dumb things he says).
There is no magical threshold, retard. It's a continuous function that grows as a logarithm.
You are the ideologically demented retard, you literally don't even understand mathematics, yet you want to talk about computation.
this argument is more nonsensical than the argument that "asi and agi cannot ever be created, because muh charts n muh graphs". asi and agi is threat regardless if a random person uses tech or not.
probably the same people in the 80s saying real time raytracing could never happen are saying that asi is not a threat and could never happen.
Friendly reminder that back in the day "the experts" said that due to muh physics and muh science that wright bros could never make an airplane.
>another false analogy that isn't even true
The theoretical foundation of flight was formulated decades before the write brothers
You can't argue with actual mathematics nor physics because you don't actually know what you're talking about
Ah but which experts are "the experts" who were correct and which experts are "le muh graphs and ideology mean x is impossibu-experts". Which one would say, "you don't understand the basedance, you can't make a flying machine because god doesn't want you too and nothing is boyant in air, the science says it's impossible.". You, you would be saying that. Because that's what you are saying now.
You are severely retarded and have a very flimsy grasp on reality.