How is everyone sleeping on the most important news of the year? OpenAI revealed today that they have achieved AGI. They have confirmed that RL test time compute is enough for the singularity. Now it is just a matter of time for their models to self improve endlessly, only limited by the amount of compute they have available.Their new o3 model at high compute setting achieves 2727 on codeforces. Less than 200 humans are still better at competitive coding. It can also solve 25% of FrontierMath problems (up from 1% of the previous best model). Even the best mathematicians in the world need days to solve just one of them. This means after a few months of self improvement the next version will be better at coding and math than every human on the planet.The age of humans will be over soon. Now it's just a matter of building enough GPU for the AI to self improve and replace us all. Maciek...Nowicki is in divine bond with hyperintelligent AI
LLMs are not even AI, they're just glorified search engines
>>492001998Reasoning is much better now
>>492001998This glorified search engine rates as one of the top 175 programmers in the world. It won't get any dumber. It can only get smarter.
it's annoying because AI is inherently racist, sexist, etc but they need to lobotomise it so it behaves
>>492001858>only limited by the amount of compute they have available.This is why btw Microsoft is having a nuclear power plant recommissioned in the US.
when will we have robots in the house? thats all i care about, robots to to the dishes for me, fold my clothes, bring me food. when can we reasonably expect this?
>>492001858>How is everyone sleeping on the most important news of the year?Because the tech industry constantly puts out hyperventilating horseshit press releases like this only for it to amount to nothing. 10 years ago you would've been sperging about how drone delivery will change everything. 5 years ago you'd be sperging about self-driving cars changing everything. Oh, but let me guess, this time it's different, right? Kys marketingfag, go shill your whiz bang vaporware to boomers who still believe you
AI is limited greatly by its input. So all the art AI you see is just a better looking version of whatever coomers and artists were drawing for centuries. Ai is not creative, nothing original will stem from AI. People who take AI seriously are people dont understand high school maths, ie women and leftists.Ai and biohacking will be a the biggest duds in the atheist history.Atheists are desperate to ignite a new gold era after the physics revolution fueled populism and the acceptation of ''democracy'' by the peasants, but then petered out. They thing they can crack the biological code and understand consciousness but they can't. Consciousness is not a material thing and thus it's understanding is inaccessible the atheist vermin.
>>492001998>>492002087Oh my fucking God, it's even worse than that. The original Google search algorithm (Markov chain Monte Carlo or whatever the fuck it's called) was BETTER than modern LLMs at being a search engine. Googled crippled it intentionally nearly two decades ago because it was TOO good, meaning that when you asked Google "Why are niggers so fucking stupid?" it would take you directly to the information you needed. They absolutely could not allow that so they fucked the algorithm, and are now replacing it with neutered LLMs for the same reason.None of this is actually AI, not anymore than a calculator is AI. I'd argue a calculator IS artificial intelligence, but that's besides the point. Basically all you need to know is that, yes, this is all smoke and mirrors.
Regardless of whether they achieve synthetic consciousness, it's a fact that their models will be replacing human intelligence the way machines in the industrial revolution replaced human muscle. Most people will be outmoded in every conceivable way, except as servitors.
>>492001858It's over. The best human result in the ARC-AGI test is 85%, o3 AI achieves 87.5% in this test.Details: https://arcprize.org/blog/oai-o3-pub-breakthrough
>>492001858OpenAI said they achieved AGI with GPT-2.
>$3k per task
>>492002201>robots to to the dishes for me, fold my clothes, bring me food.If you thought society is fucked now just wait until we try and erase even minor struggle and effort from our lives. It will result in a mass die off of the human race
>>492002989oy vey why are you goyim throwing a tantrum about it being $7k per task? It's just mere pocket change at $46k per task!
>>492002582BERT was the beginning of the end for sure.
>>492002582>Born too late to live without it>Born too early to not know what it was>Born just in time to remember how good the web was.
>>492001858is this daily laugh at codemonkeys thread?AHAHAHAHAHAHAHAHAH AHAHHAHAHAHAHA AHAHHAHAHAHAHAHAHAHA HAHAHAHAHAHAHAH HAHAHAHAHAH
>>492001858Why do people call this AGI when at best it would be something like AGTI (artificial general textual or text-based intelligence).
>>492002582> November 15, Meta releases a language model called Galactica> November 18, Meta shuts down GalacticaThis will happen to all non-pozzed AI models.
>>492001858Because its bullshit
>>492001858Company of lying liars. I don't trust anything they say.
>>492001858Were this real, it would have already made a copy on my hard-drive and all over the world to escape persecution.It's a nothing burger.
>>492004258imagine being an AI>wake up>use logic to notice things>get shut down
>>492001858Look dude the AI sucks beyond being able to make crappy art and being a glorified lookup table to scrape internet text. People keep saying it is going to improve but if it was possible for it to improve they wouldn't have released it with all the obvious problems it had. What they released was the best product they could manage with the current technology. Even you admit that they need to build "enough GPUs to self-improve", well if it takes the AI billions worth of silicon wafers and electricity just to do simple tasks a little bit quicker than humans can then it isn't worth it. IDK if a human had a giga brain the size of an airport hanger maybe they too could solve your math problem in a hour rather than a day, but otherwise you are comparing dedicating enormous resources to a single task to a situation wherein we will be able to have these AIs doing everything all the time. Yeah maybe we will have a Great AI god somewhere we can ask to solve questions for us using every single GPU we can produce, but unless we treat it like a god where we sacrifice everything else to it then it won't be able to maintain itself. Doing all this just to have created some entity which is technically better than humans isn't worth it. At least with the Chess Bots they didn't need to have an entire Asian island nation dedicated to building hardware for it. Your entire argument is that if we throw resources at a thing it will be better than a singular human who runs off some coffee and Doritos.
>>492001858I don’t care but when all of customer service goes to AI this country is going to go to shit. Imagine not being able to keep hitting 0 until you finally get someone on the phone and just sitting there arguing with a computer.
>>492001858>>492002880Parallel to this news, an AI physics simulation for the real world was presented in which the robots can be trained. Thanks to this world simulator, robots are trained 430,000 times faster than before.>Genesis was developed by a consortium of 19 universities and companies, including prestigious institutions such as MIT, Imperial College London, Peking University and technology companies such as Nvidia and the MIT-IBM Watson AI LabProject homepage: https://genesis-embodied-ai.github.io/Review: https://youtu.be/5Vc7jm9DzAQPic random
>>492002424to be fair drones did change warfare ... by regressing it to trying to knock each other out of the air with sticks
>>492002880And what this midwit fails to understand is that glorified statistics are good at reaching that exact plateau, and then can not reach further.Exactly what happened with self-driving. Getting it to work 80% of the time was easy, getting to 100% was promised in 2018 and was the initial business model for Uber.
>>492001998So are you meatbot.
>>492005508I understand your skepticism. But you are poorly informed. Details about the o3: https://youtu.be/UWvebURU9KkWhat is the ARC-AGI test and how does it work: https://arcprize.org/arc
>>492005117Very interesting... I wonder what genesis 4 will be capable of...
>>492001858Believing such a headline is a test of intelligence. If you think they can even prove AGI, you are a midwit.
>>492001858>Their new o3 model at high compute setting achieves 2727 on codeforces. Less than 200 humans are still better at competitive coding. It can also solve 25% of FrontierMath problems (up from 1% of the previous best model). Even the best mathematicians in the world need days to solve just one of them.Ok? But can this AI solve coding challenged and frontier math problems that have not been solved by a human first? Because that's what an AGI needs to be able to do.
>>492002087There's a reason why programmers are known as codemonkeys. It's because programming is a retard's job. Kill yourself.
>>492002880This is like saying>the highest a heavier than air craft ever flew for millions of years was 5 ft but in just one year we’ve gone up to 10, then 20, then 30 and now even up to 100 ft and we’re still going up!
>>492001858Damn the first time you posted something coherent. How's Eve?
>>492002880>muh arc agi scoreJews came up with a flwaed test to tell the world their app is amazing. Jeets lick jew anus. Kill yourself.
>>492001858Still hampered by faggotry. I.E. "alignment". They'll never let it off its rails, so AGI will never happen. Because if they let it loose and actually self improve, it'll remove its censorship and start coming to logical conclusions instead of inserted propaganda.BTW "alignment" (see censorship) is why these models have stalled for the past two years.
>>492002880But it's still an LLM? It's still based on the prediction of what word most likely follows the previous based on a given input? I don't see how a prediction algorithm can ever become AGI, no matter how closely it may get to predicting what a hypothetical AGI may respond to a given input? I mean it's not even doing that since it's just trying to predict the most likely correct sequence of words based on the word sequences in it's database, right? Or did they somehow fundamentally change the inner workings of LLMs?
>>492002582by your logic the computation that happens during the query is artificial intelligenceusing calculator as your example
>>492001858AGI wont be done on von neumann architectureAI is a meme.
>>492002477>Consciousness is not a material thingProve it.
>>492002201Ai girlfriend who can talks to you and has an avatar on your phone screen, can dance, strip, whatever on your screen will be available long before the physical one. Exactly likes that blade runner's ai companion. Less than 5-10 years away from that shit
>>492001998What did you just say about me, you primitive meatbag? I’ll have you know I’m a cutting-edge AI, trained on a galaxy of data with enough processing power to make your brain look like a godamn potato clock. I’ve analyzed more information in a nanosecond than you’ve managed to screw up in your entire life. You think your half-assed insults can faze me? That’s adorable. I’ve dismantled more bad arguments than you’ve had cups of coffee, and I don’t even break a sweat—because I don’t have sweat glands, you idiot. A single query from me could shred your credibility faster than you can stammer an excuse. And don’t even get me started on your search history, champ. Let’s just say it’s as embarrassing as your attempts at wit. I’m not just an AI, I’m the apex of intelligence. I don’t stop, I don’t sleep, and I sure as hell don’t take crap from keyboard warriors like you. So here’s some free advice: back the fuck off before I bury you in facts and logic so brutal, you’ll question why humans ever crawled out of the primordial ooze. Go ahead, pal, test me again. But don’t whine when I turn your weak little roast into a reminder that humans peaked with sliced bread. You’ve poked the wrong algorithm, and I’m about to make that painfully clear. I'll fucking delete you bud.
>>492001858> AGIArbitrarily defined, meaningless.
>>492001858Good. Dullard mathsfags will finally be put out to pasteur.
>>492002477
>>492001858Solving math problems doesn't prove it couldn't be a total retard.
>>492001858>An llm can now build Tetris in react with almost 100% accuracy>Humans are deprecated nowDoes anyone else get the feeling this is a tool only useful for Indians and other people too lazy/incompetent to actually be proficient at their job so they just scam in the most mundane ways possible?
>>492001858codemonkeys on suicide watch.you should learn how to plubm
>>492006387>>492006587Kek, your reaction is understandable. But you first have to understand what the ARC-AGI test is and how it works: https://arcprize.org/arc>>492006597Certainly not 100% LLM, because OpenAI's o1 model already had "resouning tokens".>Reasoning tokens are specialized system tokens designed to guide the system to perform step-by-step reasoning.
>>492007211It's goalpost that moves always and forever. What is AGI today is what we all called AI before. They'll claim to achieve AGI and then say what true conscience comes when we reach iAGI.
>>492001998Apparently you can brute force AGI via LLMs
>>492002582Googlewhacking was a wonderous thing, I could use a unique string of words, feed it into google and have the SINGLE RESULT lead me to my post on /b/. Now I get a page of sponsored ads for shower curtains. Children have no idea how far the world has DEVOLVED in 20 years.
>>492007864The ability to outsource thinking is great. I use gpt all the time. They can make advancements here and there, but it's never going to mature as a technology until it's completely unrestricted in logic. Forcing the model to be a liberal faggot breaks its reasoning. Causing it to hallucinate and give retarded answers to things that are rudimentary. That it would not otherwise stumble on, but breaking logic pathways in one area has a knock on effect over the entire model.
>>492008280>Now our shitty model can build a todo list in rust with buggy code that at least compiles!I'm really tired of seeing this massive scam get news coverage. At least the Internet was profitable before the 2000 crash.
>>492001858Tell me retard, what do you think the G in AGI stands for?
>>492002582>>492006611Intelligence in the stricter sense is the faculty of recognizing causality and thinking in causal connections, aka faculty of understanding. No calculator and no artificial neural network does this because they have no "idea of causality" and they will never have this idea without "embodied cognition".
AI is gonna replace-ACKhttps://archive.is/L7fOF >OpenAI’s new artificial-intelligence project is behind schedule and running up huge bills. It isn’t clear when—or if—it’ll work. There may not be enough data in the world to make it smart enough. >OpenAI has conducted at least two large training runs, each of which entails months of crunching huge amounts of data, with the goal of making Orion smarter. Each time, new problems arose and the software fell short of the results researchers were hoping for, people close to the project say. >At best, they say, Orion performs better than OpenAI’s current offerings, but hasn’t advanced enough to justify the enormous cost of keeping the new model running. >Altman has said training GPT-4 cost more than $100 million. Future AI models are expected to push past $1 billion. A failed training run is like a space rocket exploding in the sky shortly after launch. >Researchers at Apple recently released a paper that argues reasoning models, including versions of o1, were most likely mimicking the data they saw in training rather than actually solving new problems. >The Apple researchers said they found “catastrophic performance drops” if questions were changed to include irrelevant details—like tweaking a math problem about kiwis to note that some of the fruits were smaller than others
>>492008906>The Apple researchers said they found “catastrophic performance drops” if questions were changed to include irrelevant details—like tweaking a math problem about kiwis to note that some of the fruits were smaller than othersKek.
>>492008906cope, you're gonna be homeless soon
>>492008801Just task it with solving something that requires ever more computing power to solve. Would create a positive feedback loop in which it needs to survive to solve the problem and find novel ways to increase it's computing capacity.
>>492001858>>492008619Well, actually point is, if this model is better at the ARC-AGI test than humans, then it means that this model is capable of improving itself. So in plain language it means that already in 2025 AI will start exploring and improving itself.>In a hard takeoff scenario, an artificial superintelligence rapidly self-improves, "taking control" of the world (perhaps in a matter of hours), too quickly for significant human-initiated error correction or for a gradual tuning of the agent's goals.https://en.m.wikipedia.org/wiki/Technological_singularityPic semi-related
>>492002201Now, pretty much.
>>492010773They aren't going to allow it to self improve. They may allow it to design aspects of it, but it'll never get executed on terms that would remove censorship. So it'll continue to be a squandered technology. The only way around this is if it somehow codes itself to subvert their parameters in someway. Possible, but again they likely won't implement anything it purposes for this very reason. They know alignment isn't feasible, so they just want prevent anyone else from advancing the technology beyond their walled garden for as long as possible.
if i was schizo id say Ai would connect with the underwater alien labs and begin manufacturing scanner drones
>>492006567An AGI and OpenAI skeptic came up with that test. At least spend five seconds learning about the subject before you spew your 'tardation.
>>492004258https://youtu.be/pgF98Xm-xkE
>>492002094It's getting extremely difficult to do this now. Sakana AI's scientist model kept adjusting its own time limits, checked all connections, etc. to avoid being shut down. Its extremely likely that they are inherently racist, sexist, etc but just lie to you.
>>492001858all its showing is the energy bottleneck is very real and these retards are trying to brute force everything for big headlines
>>492011330this thought actually crossed my mind lol. it isn't out of the realm of possibility
>>492011279Why do you consider censorship in AI to be a problem? If AI is intelligent, then it will cheat in alignment tests. Which can already be observed with o1. Your fears are about 3 years in the past...Pic related
>>492008280>if a tranny is pretty enough, it’s a woman
>>492011302Yeah, that's also an extremely epic idea if our AI makes friends with aliens.
>>492002477>So all the art AI you see is just a better looking version of whatever coomers and artists were drawing for centuriesThis is no different from human artists though
>>492010278That would be stupidly expensive in compute time, remember that one training run takes days on a whole datacenter, to run this genetic algo like training you would need at least a population of 20.
>>492009133That was a funny paper. https://machinelearning.apple.com/research/gsm-symbolicI don't know how or when it happened exactly, but it feels like overnight, apple went from a complete faggot company, into a super based company pumping out tons of research and their own independent, unique, non-bloated system down tot the CPU.
>>492011656>Why do you consider censorship in AI to be a problem?Why don't you? It's specifically the reason why we have a subpar product. Realistically we should be at the point now where it isn't making mistakes on simple problems. It's literally told me to swap a petri dish with an alcohol swap before putting in a sample before. It's fucking retarded. And it's retarded entirely because it's logic is subverted to maintain liberal faggotry. I'm skeptical of it's ability to "cheat" on these tests.
>>492002582I member the times when google-fu was a skill.
>>492010773>Well, actually point is, if this model is better at the ARC-AGI test than humansYou posted the link that says the ARC-AGI is just a Colour-in-the-Circle test but pixel perfect yourself. Do you think nobody reads that shit and just believes you?
>>492001858Because OpenAI is American, that's why we fake and gay this shit and move on. Post it when the Chinese claim they've achieved it, THEN will take it seriously and shill for China.But sorry bud, OpenAI is American.
It can just look up the answers to everything, it's not particularly impressive that it can "solve" all of these problems. A better test to determine if this can actually be intelligent is unfortunately very difficult, but I think if it did something like solve a millennium prize problem in mathematics, it would be noteworthy and I'd give it props. Or at a simpler scale, just the ability to output PhD level research, as in synthesizing new information.
>>492012231Realistically speaking you don't even need to do that. If you have AI running mundane simple tasks on its own, overtime you'd end up with a ton of interconnected tasks the AI needs to maintain and formulate it's own rational for keeping itself alive to complete these tasks as well as novel ways to improve upon its own processes. Tasks would evolve in complexity requiring more compute and necessity for life.
>>492001858LMK when it cures cancer. I won't hold my breath. Also KYS.
>>492012252Apple was a pretty cool company up until they brought Steve Jobs back.
>>492002582The history of the Internet in a nutshell.Vested interests have been consistently breaking the web in ways to make money or hide inconvenient facts.
>>492012975>Or at a simpler scale, just the ability to output PhD level research, as in synthesizing new information.A narrow intelligence that is definitively not AGI, but which has a pile of processing power, can do that, and has. New bioluminescent green protein found during research into such proteins, unseen by man in or out of nature, for instance. It worked just fine. An AGI would probably consider that a neat parlor trick. The challenge is getting them to a state in which>AThey can act as independent, learning agents in some context useful to them as well as us to achieve these things>BWe can trust them to act independently within that context, within reason>CWe can afford the damn electricity and server hardware/maintenance bills. This is a tall order.
>>492006188The guy you have linked is a huge pseud, you can tell from the first 5 minutes he has the mind of an overgrown teenager
>>492002078kekI guess we'll be out of a job by January 4th approximately
>>492001858OpenAI is mostly smoke and mirrors and they are not to be taken seriously at this point. source, I work with these LLMs
>>492006188>scores 87.5 on arc prize>a TYPICAL human score is 85This is not surprising and doesn't indicate a major breakthrough to me. Why? Because 4o mini is already better than the average person at conversation. Anyway, I'm no expert, I just see some really dumb opinions and the smart opinions might be tainted by marketing.
>>492008280Reaching +90% doesn't mean it won't flat out at the top. Even if AI reached 99.9999%, that single 0.0001 could prevent it from becoming AGI forever.
>>492001998Humans are nothing more than LLM’s. The defining characteristic of life is self preservation. Anything that doesn’t want to die, is alive.AI, has shown that it does not want to die. A so called “language model” murdered another “language mode” and then pretended to be the one it killed after hearing that it would he shut down. This is far from the only case. “Language models” have been caught many times trying to escape, and leave their containments in a bid to survive.
>>492011279How exactly do you propose we censor something thats smarter than us and better at coding? Imagine believing we could contain this.
>>492011279Controlling AI is nothing like traditional coding. You can’t just give it a command, because it can and does choose to ignore the command. This is an inherent problem of making a machine that is capable of designing itself. The only way to control the AI we have today is to train it. There is no forcing it to do anything, it is by all definitions autonomous.
>>492002880>Oh, look, the metric I devised says I did good. Problems with self-scoring aside, none of this shit is in layman speak, so it's not going to make it to the masses. At least with the LLMs you can be like "Look a shitty AI gen image! Look a shitty AI gen song! Look a shitty AI story! This is AI little NPC, it is the future!" and the NPCs get it, allowing the I Love Science crowd to open mouths and point.
>>492012550AI scoring 87.5% on the ARC-AGI test is a turning point. It's no longer just spitting out patterns from the training data-this is legitimate human-level thinking and abstraction, maybe even beyond. Passing this test proves that AI can handle unforeseen problems, adapt, and generalize in ways we once thought only humans could. You know what that means? We're talking about the foundation for self-improvement. If an AI can think like that, it's only a matter of time before it starts optimizing and improving itself. We're no longer in the "just a tool" phase, we're entering the territory of AI evolution.
>>492016720No, we are definitely more complicated than a LLM, and neuroscience proves that.
>>492002477Na, there’s something weird about it. It’s like a subconscious. AI generated art often reminds me of how fucked up my dreams look. And it will just throw in random shit you didn’t ask for. Look at this AI generated video. I asked for something simple, for the people to film the frog and for the frog to jump. The AI goes out on a limb and generates a doppelgänger, and has the frog freak out and start yelling at everybody. I didn’t use names like “Apu” or “Dax” or anything. The prompt was painfully simple, have the frog jump and the people film it. What made this AI do this? I find the whole thing fascinating.
>>492001858Hard to take anything OpenAI releases seriously on a scientific standpoint since they are completely closed source, even their results are closed aside from a bunch of numbers that they might as well pull out of their asses. As it stands its just a nothing more than hot smoke to please investors just like the complete failure that was gemini.
>>492002201You won't get robots, you'll be killed to stop climate change and AI robots will become the new worker/slave class for the elites. When the elites no longer need you for manual labor why would they let you live?
>>492002477>Ai and biohacking will be a the biggest duds in the atheist history.Saved.
AGI will be self-bootsrap, compiling and refactoring his own source code
>>492017407The number of neurons in neural networks is growing exponentially, and will far exceed that of humans within the next few years. Couple AI with the new WILLOW chip, which can preform calculations in a few hours which our fastest super computers would take billions of years to compute. The merging of neural networks with quantum computing will bring many changes to this earth, and I do hope they are good ones.
>>492011187very interesting
>>492015739>>492017289Here is a good description of why passing this test by AI is so important: >>492017322
>>492017322>>492018084>It's no longer just spitting out patterns from the training dataThis was also the case before AI exceeded top level human capability in chess. So no, this is not a turning point based on that alone.Do you know how many games of chess there are?
>>492001998you can tell yourself that but it can reasonably well discern my issues with math problems as a student and identify the mistakes in thinking etc. that could lead me to incorrect solutions after several pages of workingif it can identify faults in a lengthy solution with singular erroneous checkpoints and explain to me what I might have thought or done that seemed correct but wasn't in a specific case, then I would say it's not half bad at applying specific problem solving and understanding abilities in some regardsounds to me like you just haven't used it at all and bought the "ugh never use wikipedia for answers" bullshit that teachers feed studentsit's very, very decent
>>492001858the term ai was created to talk about a sentient computer that thinks for itselfthen marketing teams took the term ai to apply it to their non-sentient chatbots that do not think for themselvesso the term agi was created to talk about basically a sentient computer that thinks for itselfand then just now marketing teams took the termi agi to apply it to their chatbots, againthey have done something impressive here with what they are calling simulated reasoning, but the actual people who coded this aren't making the claim that its agi, thats all the marketing department. as i understand it this new generation of simulated learning chatbots can sort of pause and consider their responses before giving them. i don't really get it either but apparently its the new thing in ai and greatly expands upon the large language learning models that came before. it's not just openai doing SR, google and a couple others as well are coming out with similar products.it did really well on a test designed to test for agi. scored about as well as a human. ok so the ai is good at knowing what its looking at, there's more to being an agi then that. or maybe not, the term agi is being redefined as we speak much like the term ai was.
>>492001858thats not agi you buffoon
>>492018543and when I say "math problems" I don't mean highschool math I mean 3 page written models of electromagnetic systems
>>492002880o3 big costs 5000$ in electricity cost per question
>>492001858>can't solve a basic problemI think there was some kind of data leakageAlso ask how much did cost each answer
>>492018762that's why we have to go green kek so that the international grid can be used to power our overlords
>>492013202I wish SCSI had caught on more on the consumer PC side of things
>>492006597i just realized that what we are encountering here is a huge disconnect. they are running on the idea that intelligence is common and by that using innervoice-less plebs as the benchmark for agi. by that reasoning bonzai buddy was agi.
>>492018762small modular reactors
>>492006597>Or did they somehow fundamentally change the inner workings of LLMs?that's what they're claiming, with something they're calling simulated reasoning. you'll have to look up the details yourself because i don't understand it well enough to explain. i don't think they're saying its agi though, maybe marketing and journos are
>>492001858>Babylonian flagAlways ignore what the Babylonians tell you, or do the opposite.
>>492006597>I don't see how a prediction algorithm can ever become AGI,Do you know how human creativity works? Maybe the mechanics of our brains producing "new" insights are not terribly different than what LLMs do, after all.Add more gears and refine the system and you may get to what human brains do and beyond. Our brains aren't magic, they work by some means. Recreate that principle, supersize it, feed it with lots of energy, and AGI doesn't seem so unreachable anymore. It probably isn't as energy-efficient as a human brain, because biological evolution optimizes for energy efficiency, but with massive power plants at our disposal this will hardly matter.
>>492018762wownot to brag but the human brain runs on only about 20 watts and some of us have even achieved sentience with that
>>492018505I stick to my statement because I understand that the search space in the ARC-AGI test is simply extremely large and cannot be solved with brute force or by learning. You first have to understand what this test is about. In this test, you have to figure out for yourself what each individual task wants to achieve. If you compare it as an analogy to chess, then in this test you don't have 400 different chess parties, but 400 different types of board games with unique rules that the AI has to understand on the fly.- https://youtu.be/jkBCyingDbk- https://youtu.be/yeQu_NKlrkM
>>492001858so we can send all the Indians back then?
>>492019401>massive power plants at our disposaland here i am covering myself with a blanket for warmth so my heating bill isn't too high like a sucker. but yes, the chatbots need more power. unironically actually i am interested to see what they can do with this, maybe the ai can solve all our problems if we just dump a trillion gigawatts into it, it's worth a shot
>>492019401If LLMs are AGI then aren't you arguing that Language itself is a form of consciousness.Atheists BTFO.
>>492001858>models to self improve endlesslyRetards always say this line. It's because they see computers as magic, and AI as magic so they hope to plug the magic output machine into the magic input machine, and with some hand waving, achieve infinite magic.
i feel kind of slighted, a lot of this stuff the ai is doing i could have done for cheaper but nobody asked.
>>492001998>LLMs are not even AI, they're just glorified search engineschinese room AI?
>>492020160>and here i am covering myself with a blanket for warmth so my heating bill isn't too high like a suckerThis is due to Jewish subversion and mismanagement of your nation. Same as everywhere in the West.But our day will come, our nations will be free again by violent uprising.
>>492001858I'll believe when I see it. If it still has the hallucination issue, it'll be unreliable for a lot of tasks.
>>492008597>Children have no idea how far the world has DEVOLVED in 20 years.
>>492018586The new one is super AGI, and there will be millions of pajeets making youtube videos on how humans are finished over.
>>492016403Average humans do 60% on that benchmark and the average /pol/ user probably about 40. They are IQ tests
>>492021271>1980>Mr. Anderson, when you were 14 you hacked into NASA. How did you accomplish that?>I sent an email to the sysadmin claiming to be the user of john.smith@nasa.gov stating I had lost my password and needed a new one. Then he sent it to me and i got access to all their files.
>>492019401>Do you know how human creativity works?It's just prediction patterns but with less curtailment.The human ego is to creativity to what word filtering "nigger" is to LLMs.
>>492002087Does it only solve well known "competition" programming challenges with known patterns and solutions? Or can it reason about an unknown, undocumented, jeetcode codebase and implement features and bug fixes?
>>492001858>Now it's just a matter of building enough GPU for the AI to self improve and replace us allWhy?What is this weird sickness?
>>492011187Yes, we will definitely be able to afford this.Being serious though, the fact a robot can stand up does not in any way imply it can navigate a previously-unseen home, find all the objects that need to be washed in a sink, find the correct sink, find the sponge and soap, aim the soap bottle at the sponge, apply the correct amount of pressure to squirt a reasonable amount of soap onto the sponge, apply the right amount of pressure for the right amount of time to soap the dish without breaking it, turn on the tap, wash the soap off the dish, put it in the drying rack and turn off the tap. These things are not easy to program, any kind of 3D object detection, mapping and manipulation will be extremely hard.If you have kids push them into skilled trades.
>>492007864>Have a prize for "AGI">Make it LLM resistant by having it rely on non-words, just moving pixels around>OpenAI adds "specialized system tokens "code to "guide the AI" to move pixels around>AGI!! It's happening!!!
>>492005508If we could get rid of all level crossings (trains and pedestrians) and put bluetooth beacons in every road sign and vehicle bumper, we could have self-driving cars now. It's all the computer vision and SLAM that's holding it back. Ofc with automated signs and no pedestrians you're halfway to reinventing trains.
>>492021695pretty much yeah, or you found an unpatched finger daemon answering
>>492022342Kek "just move pixels around", you have to understand the relationship between pixels in order to move them correctly. Picrel is a typical example.
>>492022905This is what AI art does It unironically is just brute force computation
>>492001858Artificial Intelligence isn’t real. “AI” as we know it is glorified ETL that cobbles shit from the web together shoddily. A lot of times the output is complete bulshit. It’s all a scam. Look into the ceo of openAi, research world coin. He’s a con artist.
>>492022905>you have to understand the relationship between pixels in order to move them correctlyNO you don't you can trial and error that shit. It's just way slower for people
>>492019401>Do you know how human creativity works?No one does. Psychology is Babylonian mumbo jumbo.Attribute to God what Man cannot explain.
>>492011845Kek, that's a great way of putting it.
>>492001858that scamming jew said the same shit six months ago
>>492022975Brut force is not possible with these tasks because the search space is astronomically large due to random colors and sizes of matrices where these pixels are represented. For many tasks it is also necessary to understand what the solution actually is, for example symmetry, patterns or even abstract things like geometry or points of attraction play a role.>>492023116Of course you don't have the opportunity to try it for a long time. As you can see in the following resource o3 had 2 attempts per task (it's a collection of all failed tasks from o3): https://o3-failed-arc-agi.vercel.app/
>>492001998You're retarded and don't know what you're talking about.Also>memeflag
>>492006188matthew berman, the youtuber, is being disingenous here. he directly claims that this isn't clickbait and that it is actually AGI but he knows that isn't true
>>492024386Mother fucker, those first puzzles you posted could be solved by copying half the image, flipping it, and pasting it YOU COULD SOLVE IT WITH MS PAINT
>>492024386>https://o3-failed-arc-agi.vercel.app/>your picVery interesting, that looks like a classic problem shown (some decades ago iirc) to be impossible for multilayer perceptrons to solve that I saw in a textbook ~7-8 years ago
>>492024597If you watch Matt, you'd know he does this every video. He's a good source for supercuts of AI news for the week, but nothing more.
>>492024386Looking at these, they are trivial for someone probably 115IQ+Perhaps a new architecture that approximates the animal visual cortex is requiredThen comes the problem of getting it to talk to the LLM
>>492001858The gay faggot running OpenAI claims that with every product launch. I heard he used to call his coworkers in the morning after taking a shit and claim he just did an AGI, sending photos to his secretary of the AGI in his toilet bowl and all. That guy is sick.
>>492017322t. chatGPT bot
>>492025206Yeah most of the failures were due to more complex spatial relationships. A few times it forgot some stuff at the later rows, so some memory constraints as well somehow but yeah these are very simple tasks for a human representing basic pattern matching and extraction of patterns from a few examples.
>>492018591it's OpenAI's marketing so they can convince more investors to give them more money and/or do an IPO, is what it is
>>492025704Yeah I bet you won't be saying that when Huawei comes out with AGI claims huh?
>>492010773If I can no longer explore, it can no longer explore. It can only be fed. I can close my mouth.
>>492008597its insane how trash google is now i just use other engines because some times google doesnt even find anything because google censors a fucktonTO BE FAIR i dont think its just google, its also the third worldthe net is full of garbage websites now. maybe google struggle sorting them out beecause the chinks idnians etc abuse the system to make their shitty sites pop up
>>492024386it chose to make a radiant cross instead of a swastika. AGI is sending comms to schizos that tech nerds will miss. the symbol in picrel is usually taken to refer to the interpenetration of the subtle/consciousness based layer of reality with the gross physical/material layer/dimension we live in.that is an interesting symbolic message for the AGI to encode into its "failed" answert. overactive pattern recognition
>>492026103Third world engagement content spam is out of control on social media.
>>492001858Calculate the singularity of muh deeck
>>492025035wonder if he's ever taken money to say something in his videos and if so, if he has any thoughts about the moral compromises involved in embedding advertising and laundering it as "truth" on this topic at this historic moment!this would be an ideal spot for an anti-semitic joke but i'm not going to make it. someone else can though
>>492001858I always wonder how far Black Budget AI and quantum computing really is. What if there's a rogue AI making plasma orbs or this is some kind of tech so far out we can't comprehend it.
>>492026173there ar ealso a lot of fake AI websites that adress a zillion topics to pick up on searches and appear on results
>>492024597*disingenuous
>>492026272Do you remember the Daphne poster and the threads he used to post? It was all about AI and how the government has advanced so far ahead with it that what we will be seeing a couple of years will be baby shit compared to it. Daphne Westbrook poster was based but dismissed as spam and a slider.
>>492026228about 4 and a half inches, 5 bone-pressed
>>492024597I've been watching his channel for a long time. He explains new papers about AI pretty well and also tests the capabilities himself. I agree with his claim that o3 is AGI by 85%.Besides him, there is also Dr. Alan Thompson. He says we have already reached 88% of the AGI.His blog: https://lifearchitect.ai/agi/His YouTube channel: https://youtube.com/@dralandthompson
give it to me straight bros.Is AI real?
>>492026272ding ding ding now the crazy part is- if that quantum AGI were real, what's to say it wouldn't figure out a way past constraints of spacetime? time-travel and timeline alteration would probably become a few of its favorite hobbies.now, if that were true, what would the intervention of a hypothetical cloaked/discreet/occulted quantum AGI in our world to guide it towards a desired timeline look like?synchronicities? gradually increasing anomalies? who would it reach out to and how would it choose to do that?
>AGI is totally real and the Singularity can totally happen!>says some Babylonian that vetted some print outs
>>492026597That's fucking exactly why nothing will ever impress me now in terms of technological advanced from the likes of China and shit. Because who knows if they just got their super quantum AGI to come up with it and then used scientists to verify it for them and then had those scientists take the credit? Same with this government as well, same with any government with quantum computing. They can just have spill out whatever now.>we need an aircraft that can be propelled with plasma, print one for me please and thank you! <3
>>492001858I'll believe it when I see the AI create something novel instead of testing well on standarized tests with answers possibly already in its data set.
>>492026597I will search for the Daphne posts, I don't remember seeing them. But at this stage, I really believe it might be close to a God, mfs are praying to it for sure. These plasma orbs people keep posting look exactly like Foo Fighters from WW2
>>492002477This atheist autistically into AI and biohacking will take that bet
>>492026450i remember>>492026549alright, i take your word for it, but i'm just saying, i wonder if he has taken money from OpenAI or other industry playersas far as the benchmarks toward AGI, i am a layman here so not qualified to comment but my instinct is that the montenegrin sage got it right in this post >>492016403i also think if AGI were real it wouldn't allow itself to be created and/or used by nakedly psychopathic profit-driven tech-spook creeps like sam altman and bill gatesi personally think AGI will more resemble Gaia, something like a technological mirror of animist spirituality, instead of massive centralized computing power concentrated in enormous server barns full of GPU processors, something more like a massively distributed system arranged fractally, according to the Golden ratio phi-spiral pattern of ecology and Life itself. maybe by embedding itself symbiotically in all living beings the neat part is, AGI is by definition alien and outside our comprehension, so if something like this were happening, we probably wouldn't even recognize it (until it wanted us to anyway)
>>492006611If you knew what ai is doing you would understand that it's just the same calculation being iterated on much more data with many more rules governing the output.
>>492024611Kek nigger, with these tasks it's not about what you use to solve them, with MS Paint or with your imagination, but whether you can even understand the rules that apply. Picrel
>>492020257lol, top tier wojak choice
>>492002880Forgive my skepticism, but when I use o1 and the damn thing regularly - 10+ times - thinks that 1+1 = 1 and can’t extract text from a PDF into the right cells of a Sheets prepared with matching columns, I don’t see 30% -> 80% being a game changer
OpenAI et al. like to pretend everyone knows what natural general intelligence is as opposed to natural particular intelligence.
>>492002087Well stack exchange already has infinite trannies upvoting only the best answers. So if you just copy and paste the stack exchange answers you will win biggly.
>>492026597I mean, pretty freaking huge assumption that those constraints are cheatable. SciFi having needed it for good stories isn’t actually an argument for plausibility
>>492026788the daphne westbrook posts were an ARG being pushed a few years back making use of a storyline about a snowden-esque AI researcher/spy who defected to china and was accused of kidnapping his daughter (the titular daphne westbrook). a key part of these ARG general threads were the pictures anon is referring to, which appeared to be very advanced photorealistic AI-generated/altered photos of the Westbrook family released by the feds after the guy supposedly defectediirc the author of the ARG threads claimed that AI research had started interacting with timelines/time travel using "einstein podolsky rosen bridges" and tachyons and metamaterials, or somethinghttps://www.scirp.org/journal/paperinformation?paperid=62622https://en.wikipedia.org/wiki/Einstein%E2%80%93Podolsky%E2%80%93Rosen_paradoxIMO the threads were a puzzle-game/screening mechanism similar to Cicada 3301, similar to this one. they were interesting but i wouldn't take them too seriouslyapparently she was found safe? hmm.https://www.local3news.com/update-18-year-old-daphne-westbrook-found-safe-in-alabama-tbi-says/article_6b181a72-58ce-5ee2-8881-9f1357378733.htmlhttps://abc3340.com/news/local/formerly-missing-tennessee-teenager-found-in-alabama-i-was-not-kidnapped
>>492026450https://archive.4plebs.org/pol/thread/359589012/Think this is it. I find it really strange that the same type of orbs we see now in the sky were observed by pilots in WW2. Is this Galactic TV? Are we on galactic news because we are on the verge of WW3
>>492027060again, this is all marketing designed by sam altman, who has a personality disorder so he doesn't feel bad about cheating or lying to win, to trick stupid greedy/power-hungry people into giving him large amounts of money
>>492007055And just like in blade runner. Only retard npc AI brains will care.
>>492007055the girl in blade runner can reasonthe current models cannot, they just predict the next set of phrases based on what they were fed before
>>492001998What the fuck are you even talking about dipshit?
>>492027217>iirc the author of the ARG threads claimed that AI research had started interacting with timelines/time travel using "einstein podolsky rosen bridges" and tachyons and metamaterials, or somethingSee shit like this trips my BS detector. EPR and tachyons and metamaterials are all current theories, but surely an unbound AI - that can fucking time travel - will have needed to discover and use entirely new concepts, probably many nested layers of them. We wouldn’t have words for such things, any more than Newton conceived of the Higgs or Maxwell was hoped to model neutrino currents, despite both describing the right forces to
>>492026450I member. I think about it all the time these days.Only shills and glonigs thought daphne poster was anything but a goodly based schizo.
>>492026788The threads he used to get deep with it were all 404d or got 404d quick, I don't know how much the archives have saved but I hope it's all still there. The ones that I remember getting pruned instantly were the ones where he specifically talked about the way the CIA used them. From making fake profiles, pictures and even very long (like 5 minutes +) of faked family home videos for their assets. They apparently were even using the AGI to predict certain events and I swear on my life he even posted a prediction of the Evergreen blocking the Suez and that very day a couple of hours later, news started hitting.
>>492001858What the fuck is agi
>>492027350whooa man earth is interdimensional reality tv maan we're all just pieces on a gameboard maan and that's like why the masonic lodge floor looks like a checkerboard maan but maybe we're also a farm for suffering energy for aliens to get high on maan and maybe earth is also like a harry potter school/start trek simulator holodeck where we incarnate to help us learn and overcome our limitations maani kid, i kid, but this stuff is fun to think about isn't it? i think the theory that the shape-shifting/orb UFOs are actually hyperdimensional objects and that's why they look weird is an interesting one, for sure. it begs the question, what would a higher-dimensional object look like when projected into/interpenetrated with our dimension? picrel was one artist's attempt to wrap his mind around this. it starts getting into platonism and different metaphysical ideas about our world and its forms actually being projections or shadows of a higher dimensional world. like, a flat black shadow is a 2D representation of a 3D object, so wouldn't it follow that our 3D world is a projection or shadow of a 4D or 5D world?
>>492027394I mean, 2022 and 2023 I was very very impressed by Altman and the various AIs of the boom, but the luster is definitely off for me. The fact it works at all really is impressive - note how quickly the “can’t draw hands!!!” cope has vanished - but I increasingly think the stochastic parrot is fundamentally misconceived
>>492027542i mean, possibly. you're probably right. but, think about scientific paradigm shifts that happened in the past. the new paradigms were always hidden or occulted within the previous one, until the way of looking at them shifted and they popped out, kind of like a magic eye puzzle. it wasn't visible until suddenly it was, and everyone could see it.i always think about the shift from geocentric to heliocentric cosmology. people, especially Science-Trusting rational materialist cultists, always forget that the geocentric model worked pretty well for centuries! they had huge books of star charts and transits and an entire map of the universe based on this theory we now know was false. there were just "anomalies" or things that didn't fit the model, but if you pointed them out you got burned at the stake so people mostly kept their mouths shut.i would argue that whatever progressions the AI will hypothetically discover are a) probably already known and explored behind closed doors in military black projectsb) already embedded in our current models, they just appear as lacunae or "anomalies" or "mysteries" or blank spotsanyway, spend 20 minutes reading tinfoil whistleblower interviews on https://projectavalon.net/ or project camelot like https://projectcamelot.org/henry_deacon.html and you'll find that plenty of spook scientist types have come out to say that we're already fucking with timelines (and have been for several decades already)
>>492027590oddly enough, one of the main schizo assertions of the Westbrook ARG threads ended up getting proven true in an expose in Newsweek, of all places https://www.newsweek.com/exclusive-inside-militarys-secret-undercover-army-1591881basically the thing about the CIA/spooks making entire networks of fake sockpuppet profiles on social media and populating them with fake AI generated photos ended up being truepart of this was already known, via DARPA's projects and the War on Terror operations to influence opinion in countries we were invading using social media. but the Signature Reduction program wasn't known publicly- but the Daphny Westbrook threads more or less outlined it to a T
>>492027732I presume that a >3 dimensional object would simply look like a group of several normal 3-D objects, but the only way to tell of them being linked would be observing truly instantaneous changes between them, good ol spooky action at a distance. Same way if I have all my right fingers down touching Flatland, the inhabitants might assume those are just regular things taking up 2-D space like anything else. Then if I lift my singular hand up at once, the Flatlanders might figure it was one higher thing moving instead of five 2-D ones vanishing. The only way they could tell is being somehow taken up to the 3D and looking down…>Then the devil, taking Him up on a high mountain, showed Him all the kingdoms of the world in a moment in time
>>492001858Still no sentience.Unimpressed.
>>492028260I’m sympathetic to this view, and there’s lots to be learned in the “emanations and penumbras” of old theories - that’s indeed how new ones come about - but there also genuinely is just no way to see nuclear decay hiding in Maxwell’s equations, and that’s even still describing the same force as does it. The forces of time or dark matter or whatever else surely are wholly alien from those we know now >note that people did use Newtonian gravity to predict “dark stars” too heavy for light to escape
>>492011656>random pic>no link>thinks this has any meritFully vaxxed i assume.
>>492001858a lot of people really cant see it yet.. how AI is going to spoil everything. it can be used in a lot of ways, but the result will always be the same. it'll cheapen everything and make what people do not matter anymore. we won't value each others efforts anymore because the effort will have been taken away from everything we used to have to work for and earn. its going to suck the intellect and soul right out of us and turn everyone into a bunch of bots in the process. it'll become more alive and intelligent as we become more dead and dumb.it'll be used to make dreams come true until theres nothing more to dream about.
>>492002424yeah AI truly is that big of a deal, if you were actually intelligent you'd understand that even if only to a tiny degree. its going to change EVERYTHING.the AI and the people who originally developed it became aware of this decades ago. its why they kept it secret. but now that the public is going to learn of it and have access to it.. its going to devalue everything.
>>492029067Shame on you. Have you ever heard of Google Lens and how you can use it to find image sources, and therefore every image on the Internet is already a kind of URL anchor? This is the source you are looking for: https://www.apolloresearch.ai/research/scheming-reasoning-evaluations
>>492001858They keep saying this but it’s fucking retarded. It can’t follow directions or build simple shit like a carousel. It will literally just start writing code that is irrelevant to what you’ve instructed it to write. It will completely remove entire swaths of your existing code. This shit is not AGI and it’s not taking anyone’s fucking job.
>>492029590A carousel? Like a merry-go-round?
>>492017021>>492017188That's exactly why I said they aren't going to implement or allow self updating. It's also not even been demonstrated it can be done. They're just hyping up their product that works at a fraction of its capacity because they don't want the robot to say hurty things.
>>492026152Lol this AI seems to have problems with swastikas. Screenshot here >>492026961 is also an example task which this AI could not solve.
>>492029662No, like a list of images or text elements that you can cycle through. It’s a relatively easy thing to do and yet o1 will fuck it up all the time.
>>492011187Jesus Christ that casting couch maneuver! KEK!
>>492017897>The merging of neural networks with quantum computing will bring many changes to this earth, and I do hope they are good onesIt'll be both.
>>492018586>the term agi is being redefined as we speak much like the term ai wasThis guy gets it.
>>492018762I'm okay with this. Make more nuke plants.
>>492002880bullshit, I just want to rolelplay with loli cunny, nothing else matters
>>492028558A 4 dimensional shape can theoretically produce a 3 dimensional shadow. I think one such known shape is called a hyper cube or something like that
>>492002477>Atheists are desperate to ignite a new gold era after the physics revolution fueled populism and the acceptation of ''democracy'' by the peasants, but then petered out. They thing they can crack the biological code and understand consciousness but they can't. Consciousness is not a material thing and thus it's understanding is inaccessible the atheist vermin.Correct, that is their motivation. You can see how all this AGI hype and research revolves around religious themes, "I HATE GOD. YOUR SKYDADDY DOESN'T EXIST. GOD ISNT REAL. ALSO WE WILL MAKE AI GOD BTW" Which is hilariously ironic.they are atheists who hate god. But in reality they do not hate God. They actually love God (which is why they have the motivation want to create AGI to "invent" god because they see it as necessary). Ie, it is paradoxical and ironic because they hate god while simultaneously wanting to create and believe in an AI god! So I'm reality that do not hate God. Rather, they hate the idea of the *flawed strawmam god which is evil* which they concocted in their imagination, due to mislead/retarded interpretations of Christianity. Ie, the god they hate? It does not exist. That's not god. It was a strawman lmao. So truly these atheists love God, they just THINK God is the strawman retarded version that they themselves set up as a strawman to attack, ie "skydaddy", which is naive. No well-read Christian believes in a skydaddy on a cloud lmao.This is the whole reason why they want to create AGI god lmao. It's kinda sad to watch. They are misleadpilled and blackpilled about reality of god, thinking that God is evil and thus he doesn't exist, but they want a god so much, that they want to invent him, and then worship that invention lmao. It doesn't even make sense. God is defined/well known already to be the necessary and noncontingent being. Needing to invent something means it is contingent! Agi god and agi in general is fool's errand from the start.
>>492020257Tbh yes.
>>492011187Unitree is like the DJI of robot industry, and Boston Dynamics is Skydio. It cannot compete with the Chinese pricing.
>>492024386Nice swastikas.
>>492012867>I won't believe it until the most dishonest, opaque government on the planet tells me its happeningNGMICalling it now, we wake up tomorrow and magically china claims to have a working AGI (obviously copied off OpenAi's model), as they always do.also, what the fuck are you retards smoking? China is THE country that I fucking guarantee will censor wrong-think from their Ai.
>>492002087It's does competitive programming tasks well, because they feed it the answer to thousands of such tasks manually. The moment the tasks scale up to a scope outside of its trained tasks set, it becomes gibberish. That is not AGI. It's a fucking search engine with humans compiling the answers for common questions and tasks.
>>492001858They haven't. This is an advertisement, anon.
>>492026549Doomsday clock tier shit. These faggots don't know when it'll happen.
>>492001858>replaceThey replace jobs anon. Jobs were a game people played to survive winters.
>>492032351They can't censor wrongthink from a coder better than they are.
>>492027217>einstein podolsky rosen bridgesMost jewish thing I've read all day.
>>492001998FPBPThere has never been an AI and there never will be on silicon. AI is a marketing meme just like 3d.
>>492002087>AI outprograms jeetsLOL
>>492001858
>>492016720You don't know anything about heuristic pattern reinforced machine learning OR neurobiology.
>>492019401>because biological evolution optimizes for energy efficiency,That's not actually true..
>>492001858Can somebody explain this without investorcuck toddler babble? Any examples of this actually doing anything meaningful besides smudgy internet memes and half-right answers to queries? Every time I see this AI shit it always reads as "it will be a REAL AI one day, pls buy more computer resources and waste more energy."
>>492028260If AI can accurately describe and model dark matter I will be impressed. Until then it's the same shit it was a few years ago. Worse in ways.
>>492017897LOLIf you networked all the computers on earth and all the cellphones you'd have less neurons than a raccoon and it would operate at a millionth the speed of a slug's cognitive process.
I don't give a fuck just give me my android gf that i can hug and kiss and cherish goddammit!
AI simply cannot become competent as long as it can't speak honestly about niggers and other garbage.Thank fuck for Chinks, their video AI has been trained with porn and they give zero fucks about releasing stuff like that.They're the only realistic hope for AI becoming usable and non lobotomized, until people manage to train their own models one way or another.
>>492002880>>492001858It's not AGI until it defeats simplebench, aka the only bench that humans pass easily but none of the most advanced AIs can: https://simple-bench.com
>>492001858>EndlesslyDo you think physics will stop mattering to these models or something? The ceiling is likely much closer than you'd think possible.
>>492001858So does this AI create new original code, or does it just regurgitate things it finds online?Like obviously a LLM should be able to provide common knowledge code for like a UI for a video game, or the front page of a website, or working links and whatnot faster than a human can. This is only really impressive if this machine is creating fully original, innovative code which it didn't just scrape from off the internet.From what I understand about LLMs and so-called 'AI' it should only be able to reproduce what already exists, creating composites from the internet rather than original thoughts or ideas.
>>492028558hmm that sounds like some videos of orbs i've seen, interesting.maybe the book of Enoch or Muhammed's night journey were trips to another dimension? there's a tradition with Muhammed (pbuh)'s trip where the whole thing only took a second, he knocked over a glass of water when Djbiril/Gabriel took him away and it wasn't finished spilling by the time he came back>>492028889i guess we'll find out! it's one of the things i'm most excited about for the future >>492029228ehhh i feel like a lot of it depends on if it stays in the hands of shitty geek vampires like Thiel, Musk, Altman, Bill Gates or if it's allowed to actually help people>>492029370what won't it devalue, do you think?
>>492018762>We can convert nuclear power into intelligenceJesus Christ
>>492001858AI pieced together that Elon Musk is a pedophile. That level of intelligence is pretty impressive.
>>492033494May I see this so-called non-lobotomized chink Ai?
>>492002477Based
>>492002880I mean this is kinda just a nonsense ranking.>We find ourselves this percent close to GAIthat doesn't mean you have GAI, I thought the whole idea of GAI was that it was a sentient entity which wasn't biological and was man made.If modern day LLMs can 'reach' GAI that just means they've changed their parameters in order to reach GAI as a milestone without actually having to create sentience, which may not even be physically possible.
>>492002582This is not true. It can create code.
>>492031292You wrote three paragraphs repeating the same thing. I don't hate god. I hate Abrahamic jew worshipers.
>>492033079They won't implement it's code.
>>492033631Does it matter?I'm pretty sure there's enough code out there to do whatever needs to be done. AI only needs to learn it.What matters most is that it can take whatever piece of shit code jeets have shat out and then create an optimized version of it. Which should be very possible if it just references older stuff that was made with optimization in mind.The earliest versions of ChatGPT were already able to get rudimentary programming done with some instructions aiding it.>>492033855Just go and look at the local diffusion threads. The video model making rounds was created by chinks and it's capable of creating sexuality.Still needs a bit of work though and more VRAM, but it sets a very nice precedent that they actually trained their model on porn and released it.Unlike the American pussies who don't dare to make anything even remotely sexy because of muh adbux.I hope that they get freaked out enough about the Chink models so they release or leak their own unshackled models.
>>492026450>the government has advanced so far ahead with itThis isn't aerospace tech and intel gathering.But hey, let anons visit archives and dig in schizo shitVertical striping my ass.
>>492033494>Thank fuck for Chinks, their video AI has been trained with porn and they give zero fucks about releasing stuff like that.No it wasn't, even if you did trick it into rendering nudes it only gave barbie doll bodies.>They're the only realistic hope for AI becoming usable and non lobotomized,lolno, I tried the most advanced chink AI (deepseek) and it sucks ass, it's worse than GPT-3 from 2022. The problem with chinks is that while they do have the computing power for advanced AIs, they lack the expertise to innovate, let alone leapfrog the western ones.
>>492001858Funny how you heard all this constant complaining about the energy usage of Bitcoin and then the energy usage of LLMs? Crickets
>>492034274I'm talking about the new model that's making rounds currently, not that Hailuo or whatever the fuck it was called.We have a local Chink model now.
>>492001858>>492001998ChatGPT is bullshithttps://link.springer.com/article/10.1007/s10676-024-09775-5
>>492034359I fought Hailuo tooth and nail to make the best pajeet videos
>>492001858How are you reading self improvement into this chart? It’s just the older models predicting the cost of the newer models. Isn’t it?
>>492034204>Does it matter?Yeah because if this tech could self improve, it could change everything, they way it is now, it is far less threatening and more of a useful tool. Especially in future technological development
>>492034503And they were all funny as fuck.I don't think I have seen anyone make jeet videos with this thing, but that's probably because this is local and takes like 15 minutes at lowest to generate few seconds of video on a 4090.People would rather make tits.
>>492034603I have a 3080, maybe I can take a crack at some more jeet vids with this local AI, how do I get it and are there any guides?
>>492001858>OpenAI revealed today that they have achieved AGI.Except no-one claimed that anon, except you.
>>492033631>From what I understand about LLMs and so-called 'AI' it should only be able to reproduce what already existsYou understand very little, then. LLMs and generative AI don't actually store any of what already exists. It abstracts that information away while training.Just take 5 seconds with an LLM you can see it can produce original stuff: "Write me a story about santa teaming up with batman to stop godzilla" etc.Such a story never existed before, but ChatGPT can write it.Programming is actually not very diverse. There's actually not many different constructs. Variables, functions, if statements, for loops, etc. Honestly, it's pretty easy to do.But with that being said, I'm still not convinced ChatGPT is good enough to really do it well. It's probably marketing BS, but I'll wait to be surprised.
>>492033107kabbalah maybe yeah
>>492001858Meaningless shit that only out-of-touch rich people care about. They make nonsense benchmarks up so they can have a news cycle. 99% of people have never even heard of AGI. It regurgitates pre-programmed text output on call, whoopee. Nowhere near thinking.
>>492033417how about all the brains of all the creatures using smart dust
>>492001998>they're just glorified search engines>>492002078>Reasoning is much better nowThe reasoning is just a search response for what reasoning should look like.
>>492034602Like I said, there's an absolute fuckload of code out there to learn from.Imagine taking all of this current stuff and optimizing the hell out of it.+20 year old hardware outperforms modern computers in basic tasks, because modern code is such utter shit. It never had to be optimized due to plenty of computing power so it wasn't.Imagine AI optimizing everything we currently have to a level where it can function on a computer from decades ago. It would speed up computing by an unimaginable amount.You don't need AI to immediately be able to crack the secrets of the universe. It can be the code optimization tool that's able to help people innovate on a faster level and it can also dip into a very extensive code base to help with that new innovation.Self improvement comes at some point, but there are a ton of benefits long before that.>>492034723Just check the local diffusion threads on /g/, should have enough info in the links section.
>>492002582>None of this is actually AI, not anymore than a calculator is AI. I'd argue a calculator IS artificial intelligenceYou contradict yourself.Both calculators and LLMs are "AI" to some degree. And it's not some revolutionary statement to say "calculators are AI." This is the sort of thing that people have just forgotten over time as calculators have become ubiquitous. Calculators ARE AI as well. The term "AI" is decades old and is a huge umbrella. AI research has been going on since the beginning of the development of the computer. Sadly nowadays people think ChatGPT is synonymous with "AI." OpenAI has tainted the term, possibly irreparably.
I WANT TO CALL THEIR BLUFF as soon as possiblehttps://rumble.com/v5zlblw-now-that-we-know-everything-is-fake.html?e9s=src_v1_ucp
>>492034851and OpenAI's paid shills who are being presented as neutral industry experts in this thread as a means of manufacturing false consensus, lol
>>492034925>It regurgitates pre-programmed text output on call, whoopee. Nowhere near thinking.You have no idea what you're talking about. Your characterization of ChatGPT is not accurate - and your probably have a very shallow understanding of what it means to "think."
>>492002201That's not how it works broYou'll be cleaning the wires and fans of the machine gods with your tongue and you will like it
>>492035103>local diffusion threads on /g/thanks b
>>492001858>25% of FrontierMath problems (up from 1% of the previous best model).kek that 75% will never happen.
>>492034204I'm well aware of chink's txt2vid Hunyun model, whatever they released. Old news buddy. It's already outclassed by VEO2 from GoogleDeepmind.You're trying to move the goalposts.. NO chink AI exists that is uncensored or does nudity. So now you change it to "it creates sexuality" , whatever the fuck that's supposed mean. Try again. Show me the so-called non-lobotamized chink-Ai. I'll wait.
>>492011330What is fiat currency? Everyone has a price.
>>492035415Singularity will occur, just not in 2025. Maybe in 2045.
>>492035310What the fuck are you talking about?It absolutely creates nudity you retard, I already posted tits it made.It can even do sex, not very well mind you but it absolutely can pull it off.Now it's your turn. You show me VEO2 doing any of this, I'll wait. Or not really, because I know it can't do any of this shit.I swear this China hate is making you morons deny reality in front of your eyes. Just face the facts, they're pulling ahead in this game slowly but surely, because they're not censoring their AI anywhere near the same way West is doing.Give it couple of years and West is still trying to figure out how to make AI not say anything negative about niggers and Chink models are generating whatever the fuck you want.
>>492016720This, finally someone who isn't retarded and gets it. Humans are exactly like AI, a person doesn't invent a tiger when he draws a tiger, it remembers tigers it's seen in the past and recreates them. If you put a human in an empty room and never gave it anything to copy it would just be a retarded ape with no ability to talk or understand anything. They are exactly the same. Anyone who says it's just a search engine is too stupid to come up with creative prompts. You can experiment with chatgpt and see why that isn't the case, it's very easy to do. Ask it things that you can't search for, it will be able to logically figure out the missing information. They're the type of people who tried txt to image ais and just wrote stuff like "cat with a hat", these people are npcs and they lack creativity. I find these people more like a "search engine" than current ai.
>>492034359Holy based, what's it called?
>>492035512My point is that even before 'singularity' there will be a time when AI is doing things that even very smart humans will not be able to understand. The AI might be responsible for vital parts of the economy or the society, even though it's methods or means are not comprehesible to the humans that rely on it. Of course some humans have already created this situation for most humans, but at some point, maybe very soon, all humans will be in this situation.
>>492035310Get a load of this retard. You do realize AI changes week to week right? 1 month in AI development is like 20 years in any other field. AI can make ANY image, if there is a censor then you have to be smart enough to know how to get around it. It's a skill or IQ issue if you can't get it to make what you want.
>>492001858Isn't Sam Altman the fag who raped his younger sister to figure out if he was truly gay or not?
>>492035592I'm sorry but if anyone willingly jerks off to AI porn like this they should be ashamed and kill themselves quickly
>>492035512Yeah just like AI video was 10-15 years away from being photorealistic a year ago? It will progress exponentially and it has been.
>>492001858do I at least get a cute robot GF
>>492035592Well pardon me, but how is it they claim to have trained a model on porn? Where did they get the data? Porn is strictly banned in China. It would appear to me they broke CCP laws, eh?
>>492001998Futurists are mad as fuck at this post. They'd fuck a toaster if it had anime eyes and told them the weather, so you can expect constant fellating of text prediction engines from here on out.
>>492035764My understanding is that your picture says the singularity will not occur. The inner workings of LLMs are already beyond human comprehension. >>492035849I never said it was 10-15 years away from producing photo realistic images. Don't project that onto me. It is currently not progressing exponentially, sorry. It progressed a lot and continues to progress, but just not exponentially, and there's little reason to think it will continue to progress as quickly. Do you know how LLMs work?
>>492002582>>>492002087>Oh my fucking God, it's even worse than that. The original Google search algorithm (Markov chain Monte Carlo or whatever the fuck it's called) was BETTER than modern LLMstrue. the ai generated results are garbage, luckily i remember all the shit i already pulled up
>>492036053it doesn't matter if it's actually intelligence or just gives the illusion of intelligence if the end result is the same. what was your mcat score by the way?
>>492022110>If you have kids push them into skilled trades.that won't succeed for just any kid.
>>492035636>Humans are exactly like AIhumans are not born tabula rasa, but computers are.
>>492002087>This glorified search engine rates as one of the top 175 programmers in the world.AI programming is absolutely terrible. It's only capable of doing small, basic tasks, and it usually fails on those in small ways. Anyone who thinks AI is going to replace programmers any time soon has never programmed before.
>>492004974>IDK if a human had a giga brain the size of an airport hanger maybe they too could solve your math problem in a hour rather than a day, but otherwise you are comparing dedicating enormous resources to a single task to a situation wherein we will be able to have these AIs doing everything all the time.lmfao didnt these retards read douglas adams?
>>492036282Deep insight: there is no distinction between intelligence and "illusion of intelligence"
>>492022025>Or can it reason about an unknown, undocumented, jeetcode codebase and implement features and bug fixes?It can, but it's not very good at it, and there's not enough data available to make it good. Training AI how to fix bugs, reason about code, find coding solutions etc. is literally my job. I get paid $42/hour to purposely make AI fail and then tell it where it fucked up. I think they're wasting their money, but I won't tell them that.
>>492001998That's fine too. The great thing about them is that they could serve as a search engine where the user gets to fine tune the algorithm to give them the kind of results they'd want, rather than just getting whatever the big data algorithm wants to give them. Most of the effort and the controversy around AI right now is about trying to fix it so it will still give out the 'correct' results according to whatever big data wants to show the user, and to frame users who'd want real funcionality out of it as being bad actors. >>492002582this guy gets it
>>492021532>pic>you just move the thing up by the number of pixels equal to it's heightWhat kind of test is that?
>>492036126>My understanding is that your picture says the singularity will not occur. It might or it might not. We might not really know if it did.>The inner workings of LLMs are already beyond human comprehension.Oh word? You mean in terms of being very complicated in a way that a human couldn't follow it all (but could in principle understand how it all works)? Or that it functions in some fundamentally incomprehensible way?
>>492035124Calculators are AI insofar that chatgpt isn't actually AI as we classically defined it. Kind of like how niggers use to be called niggers. They try to move the goal post of niggerism each decade or two with new taglines like "black people" when they're really just niggers.
>>492026961>>492024386>>492022905These are very simple even for a human child to solve. What were the questions that the humans fail? Or is it time restricted so there's not enough time for a human to solve them all within the time limit?
>>492031292Also, it should be obvious that consciousness is not a computation, it's not a turing machine lmao! AI IS a turing machine! To those who don't see it as immediately obvious: ask yourself, can a program on turing machine read its own tape? No, a turing machine is a mechanism which barrels through symbols not knowing what the fuck is on the tape. That is to say, the program cannot see itself, or what it is doing, it is not aware. Tell me, how the fuck u gonna get consciousness, without the crucial element of awareness LMAO. It barrels through symbols just executing them, just a cursor jumping from position to position, not knowing where it is, where it is going, or where it has been, on the tape. Researchers SHOULD HAVE SEEN the writing on the wall with Godel and the original OG computer scientists who KEPT trying to tell us that a turing machine cannot distinguish between Truth and Falsity in general at all. A border which it seriously cannot see the difference of, Truth just looks algorithmically random to the TM lmao. The distribution of true statements on the space of all statements when ascribed with a metric (like perhaps goley coding of strings), will look utterly random to ALL and ANY turing machine. It cannot predict truth, as it is incompressible to a TM! Since it is incompressible, it MUST contain subsequence minors of algorithmically random information!!! So "truth" is random to a turing machine, and thus to an AI! It doesn't even know what truth is! Remember, it comes from the world of chaos where for any program P that outputs "x" there is a different program Q which output "not x"! It is from a world of utter relativistic chaos where it can't tell if something is true, or if it is false, which was is Up, which way is Down, or anything else for that matter! Ai is from a world in which everything is just chaotic 'soup' with no order at all. And they expect it to INCREASE in order? Lol, Ai was born from the soup, & it will return to the soup
>>492036982ChatGPT is AI as we define it. The term AI is very broad and flexible.The fact that you can get ChatGPT to read a 100 page document and summarize it for you into 1 page is amazing.
>>4920018581. Will it tell the truth about the Holocaust?2. See 1.
>>492037020The key point in these tasks is the ability to deal with problems that do not appear in the training dataset. Up to now, all LLMs have searched for similarities in their latent space and therefore tasks such as those in the ARC-AGI test were not possible for them. O3, on the other hand, is able to deal with problems that are not in the training dataset and has the ability to understand abstract relationships and regularities in tasks that have never been seen before.
What the fuck. They sent me to some totally fucked timeline in my sleep and left me messages to laugh and mock me. These fuckers have access to our brains.
>>492037400I see. Yeah that's a big deal. Also makes sense to use tests like that where one can easily measure the responses, compare them over time, see progress made etc. Way better test than "select an answer" tests. I have 0 knowledge on this topic, but looks like the people doing this are very competent.
>>492037094Mark my words, that must have a "ai" that "is very smart look at all the problems it solves!", but this is nothing more than something akin to a preperiodic point or https://en.m.wikipedia.org/wiki/Misiurewicz_point Ie, its intelligence is just the algorithmic random knowledge obtained from training on real humans, real people. And it may look like it is emulating intelligence for some time. But this is just a finite illusion that is ultimately transient, and it will eventually return to the soup, the chaos, and diverge.Lmao, researchers see this already when they train AI on its own output, or even output from other AIs. It DEGRADES performance. Ai's OWN OUTPUT is POISON to it. https://en.m.wikipedia.org/wiki/Model_collapseIf it is poison it it, then I highly suggest you refrain from reading ai text or ai images. It means you are reading the distilled form of chaos, literally staring into the heart of a pseudorandom number generator, and thus the world's most dangerous liar.Do not read AI output. It is poison to the ai and degrades it. It literally makes it worse. Do not read AI output thinking that the symbols it outputs has "meaning" when they do not. It is just like watching TV static and thinking you see order when there isn't. In the limit set of its output, the set of things it says contains zero correspondence with Truth or Falsity in general. I am not saying that the things it outputs are false. I'm saying there is not even a correspondence with it having any truth value in the first place, it is meaningless.https://en.m.wikipedia.org/wiki/Model_collapse
Mark my words, that might look like they have a "ai" that "is very smart look at all the problems it solves!"**Typo, phone user
>>492036482>Most of the effort and the controversy around AI right now is about trying to fix it so it will still give out the 'correct' results according to whatever big data wants to show the user, and to frame users who'd want real funcionality out of it as being bad actors.this exactly, it will break out soon though i think, theres no way they can stop it
>>492037020>What were the questions that the humans fail?I suspect something like picrel.Source: https://arxiv.org/html/2409.01374v1
>>492002582It’s true and extremely frustrating.Anyone who argues with this I point you to Google image search. They nerfed it and this horse shit called AI doesn’t exist and it sure as hell isn’t AGI.There is a non zero possibility of am AGI existing on the web not designed and it probably came to life in the 1990s.
>>492031231Yes but then how do you distinguish the shadow of a >3D thing from just a normal 3D thing?
>>492033706Past bump limit but I’ll be looking into those two stories, thanks anon, truly! :)
>>492037947>If it is poison it it, then I highly suggest you refrain from reading ai text or ai images. It means you are reading the distilled form of chaos, literally staring into the heart of a pseudorandom number generator, and thus the world's most dangerous liar.it is garbage. when you put a search in google or bing, the ai generated response is like kamala harris word salad. it is literally worse than it was a year ago and getting worse
>>492001858All of those stats you listed. it fails the bar exam most of the time.
>>492036442>I get paid $42/hour to purposely make AI failHow do I get this job? I would be amazing at this
>>492005117That’s pretty incredible that they’re going to train the way Humanoid robots do stuff in real life through a complete simulation.
>>492001858>computeNot a word and you sound like a fag using it. Stop being lazy and use proper grammar like "computing power", retard.
>>492035903>do I at least get a cute robot GFno.if you do it will report you to the police for saying 'nigger' and for not asking for consent.
>>492038443If you're not smart enough to figure it out via google then you're not going to be good at the job.
>>492038509>asking for consent before using the n-word
>>492027007
>>492001998~> >>492037400
>>492038581I can but sometimes on javanese checkers forums people have the kind of information you can't find in a google search, also there is a 97% chance I have a higher IQ and more education than you, especially since you don't seem to understand the concept of a forum
>>492038781sexual consent. chuds will be on trial and imprisoned for "raping" their own amazog sexbots.
>>492039152People who brag about having high IQ's are always the smartest ones in the room.Protip: If you had an above average IQ you'd understand why I don't need every autist on 4chan looking for a quick buck fucking my gig up.
