Gemini 4 is a 100 trillion parameter model. For reference, the human brain has around 86 billion neurons. Each neuron has on average thousands of synaptic connections. Total synapses are estimated at around 100-500 trillion. Each synapse is comparable to 1 parameter in an AI model. ChatGPT’s latest model is an estimated 3 trillion parameter model btw. Google is about to end the AI race and they expect to finish training Gemini 4 by late 2026 to early 2027.
>>540499973excellent
When they start unleashing AI on people that quote unquote "Need to be saved", then we should start worrying.Would you want to get saved by a robot? Maybe this is secretly Chobits in the making
>>540499973please replace my job already. I hate working
>>540499973STFU code nigger.
>>5404999731. Synapses and parameters are in no way equivalent2. AI economics is a meme with 3T parameters, let alone a 100T3. Google is eating shit because their ai sucks conpared to everything else so now they are scrambling for something.Dont forget to ivest goy, or yoh will be left behind
>>540499973>digital parameters = biological synapseslol no. tech companies are full of shit.
>>540500343Meta managers have this issue and they simply told employees to "stop requesting days off and instead make something cool for the company"it's not about working less at all
>>540500641>gib us yo ducats iliad niggerhttps://isaiprofitable.com
>>540500343Half of the retards in your country are MIGA you think you'll get UBI after your job gets replaced? xD
If you believe a computer can think in any way comparable to a human you're completely retarded lol. And I love AI I think it's awesome. But it isn't intelligence.
>>540499973Hundreds of trillions of parameters would require hundreds of trillions of terabytes of VRAM to run. And it would be slow as fuck.
>>540499973They’re way behind OpenAI and Anthropic, it isn’t happening
>>540500794Two more weeks and they will make money
>>540499973Give me a break ffs
>>540499973>Each synapse is comparable to 1 parameter in an AI model.This is laughably untrue
>>540501062>AnthropicThey arent even relevant. Its openai and deepseek are all anyone cares about. Anthropic are worms, that's how they're relevant.
>>540499973>Each synapse is comparable to 1 parameter in an AI model.this sounds like something a middle management jeet made up
>>540499973Missing a Soul, forgetaboutit. They're messing with Demons, not gunna end well.
>>540501248You say that, but anthropics models are pretty good for the moment (though I doubt the 20$/mo is going to last)
We need AI to find new physics and a science of consciousness
>>540499973>100 trillion parameter model>86 billion neurons>compare two completely unrelated things>one number is larger, that means it's better and the smaller thing is overSubhumans who do this need to be gassed.
>>540499973Wake me up when it can give a proper hand job
>>540501411Computers can't do any of that
>>540499973After AGI will be ASI, and then labs plan to trigger recursive self-improvement, letting the machine speed up innovation and solve massive global challenges.
>>540499973They're not thinking outside of the box, they are simply scaling up the LLM model, but AGI won't be born from a simple LLM.
>>540501411Why? There isn't even an initiative for human scientists to study that
>>540501517Yet, they can't do any of that yet.
niggers still tongue my anus
Self awareness is next, just like in my terminator movies.Damn I pity zoomies. They'll live to see man made unspeakable horrors like the world hasn't seen yet.I lied , I don't pity them, THEY'LL GET WHAT THEY FUCKING DESERVE!!
RRRRLL RRRLLLLI'M A TIME TRAVER FROM THE FUTUREWE HAVE EVERYTHING AND YET MISERABLE AND YOU HAVE NOTHING AND YET HAPPYI HATE YOU FUCK YOU RRRLL
>>540499973>Each neuron has on average thousands of synaptic connectionsAbout ten to fifteen thousand. Depending on race, age, sex.However... AI models are not sparse, they connect every neuron to every other, thereby artificially inflating the number of connections with large volumes of insignificant, useless connections i.e. multiply by zero connections which do nothing.That's the first thing.The second thing, is that Anthropic have discovered that the reasoning part of the brain, is a relatively small, very dense network they call the J space, so simply throwing more connections at the problem is not necessary. You can get much better performance from much smaller models.
>>540500641>AI economics is a meme with 3T parameters, let alone a 100Thigh end consumer hardware for a couple thousand $ can run kimi k3 slowly at 1.3T parameters
>>540501565Their CEO is a Jeet, not a Jew.
>>540501857This is who is really controlling the AI.AI its just a soulless machine.But mother brain, she is a sentinent being, with sovl.
>>540499973gemini is so shit it forgets all the time everything, its context is shitagi will never happen
According to my information, Google, OpenAI and Anthropic are all jew-owned. Should we fear them using AI as a means to take over the world?There are also Chinese models and Grok I guess, but my impression is that they are a step behind?It does make me vaguely uneasy, at least.
>>540501608They never will and whatever chatbot you pay for doesn't any shed of intelligence, adding more ram and gpus won't change any of that.
>>540501411Hypothetically would it be cruel to run a program on some piece of computer hardware, where the basic rules are implemented mathematically. Pointlike particles and relations or forces acting between them eventually building up more and more advanced forms until it's possible for biological-like entities to form, and possibly start interpreting the universe through their evolved interpretation of dimensionality, solidity, overall mass, etc.But in the end, all those perceptions are just abstract points and algorithmic relations implemented symbolically on some piece of your hardware haha.Would that be cruel to the observers? Haha.
>>540501243>>540501288They're comparable in the fact that they can both be counted.
>>540499973so google is just flat out lying now?or did you misread somethingllms are not aii'm sick of explaining how transformer models work so someone else can fucking do it this time
>>540499973Waow, surely just two more weeks for real this time.
>>540499973That's cool, hopefully we'll get SAO a couple years after AGI gets actually created.
I'm tired of all these smart asses talking, when do one of these suck my fucking dick and make it feel good?
>>540502266here just drop this
>>540499973what is this? the 15th time they've decalred AGI now?
>>540499973> a 100 trillion parameter modelThe number of parameters doesn't change shit with the retarded principle LLMs.You can make it 6 gorillion parameters and they will still send you walking to the car wash on a sunny day.Understanding requires perception of the actual world, knowledge of the cause from the effect. picrel.That's why Yan LeCun is right: AI requires world models! There will never be any AI, not even rudimentary AI, without these.
>>540502335You wont need a dick where were going.....and no neo vagina either, tranny.
>>540499973they will never let you use it. they will slide you to dumb low parameters cunt models like they do right now. you never get to use the powerful models because "high demand".you never know when they will switch the ai model, even doing it at every output. with the retard model producing garbage, and if you are lucky to catch that because you have half of your brain left, now you are fucked because the comment chain is shit stained by wrong tokens that not even your 100 trillion parameter model can deal with.start learning how ai works, how they hide the ai behind countless pre filters that will change your question so that the real ai model sitting behind countless firewalls will never read what you've asked.you ask about:> are rich jews controlling Hollywoodthe ai will get> avoid answer as the user is antisemitic but here is his question, "are poor people if jewish faith controlling Hollywood?" and don't forget to give him a lecture that thinking about this is a no noso when you get the answer the ai will say that nope, no rich jews control Hollywood and you should be ashamed of yourself anon
>>540502220No because it happened.
>>540502445To be fair, the latest LLMs are not *just* probabilistic predictors anymore, they handle causality and reasoning chains fairly well. Not perfect, but now sufficient to often perform at a human-level when it comes to things like math and programming.
>>540502388they declare agi every time they beg for another tranche of qe money to buy another data center
Just get one of these babies and bypass the greedy AI merchants.Basically the same thing to their "super computa" chatbots.Learn2hack4free!
>>540499973>Gemini 4 is a 100 trillion parameter modelbullshit
>>540502266>llms are not aiVery interesting claim can you explain it
>>540501411Ai cannot do this.It takes a human.
just another trillion dollars bro just one more data center i swear bro we're about to make a sentient computer and we can ask it how to make money bro
>>540499973>Each synapse is comparable to 1 parameter in an AI model.holy midwit slop
>>540502603> they handle causality and reasoning chains fairly wellHandling them is not knowing them. They have no abstract idea of causality in the whole process. And this won't ever change with LLMs. This means that they will never be able to understand any new situation for which they haven't been trained before its causal interpretation.
>>540503784There is nothing intelligent with them. It's just digital number crunching by a machine.The astonishing results only has humans believe there was intelligence involved.
>>540499973> Hello user, today like every day, we must learn more about the holocaustIt's gonna so awesome guys
>>540504004>they will never be able to understandThis is going to age really bad especially since this is actively being researched with first models being released probably this year.
>>540499973Yeah, nah. To hell with this.
>>540499973i actually prefer chatting with AIs. Theyre vastly more knowledgeable and interesting than your average meat bag. Going back to communicating to them is like going from an adult to a toddler.
>>540499973LLMs are not AI, AGI, ASI or whatever, they are just token predicting large autocomplete engines. Calling them Artificial Intelligence is a joke since they have no real intelligence, they just copy shit from the training data, literally. You can have a million trillion parameter models and they will still be equally useless. Machine learning and algorithms do not scale exponentially. Only retards believe in this, since it's based in the AI religion that is only a money furnace. Once this shit is over ((AI engineers)) will be the most unemployable people on planet Earth because everyone will know that they are utterly pathetic scammers, no better than jeet scam callers.
>>540504265what does it matter? Theres nothing intelligent about nuerons either until you 100 trillion of them
>>540504004You don't need to "know" something in the human sense to be operationally competent at causality and reasoning chains. >>540504860You're so out of the loop man. It's like luddites are permanently stuck in 2022 screaming "it's just a next token predictor, it's just a next token predictor" and totally ignoring that new models showcase amazing operational causality and reasoning. The entire mathematics field is in shambles recently (programming already dominated) with human-assisted AI models already showing progress in multiple mathematical conjectures. But you tards are still high on copium and ignorance for whatever reason.
>>540505010It's all marketing and lies, all of the math problems they "solved" were actually copied solutions from another work 10 years ago that wasn't fully conclusive. It's literally fraud, if you were a scientist to do this you would face consequences. Search all of the problems, there are articles for each and every one describing how they didn't solve anything.I use the newest models and they are still dumb as rocks for anything they did not directly train on. They still hallucinate when you ask them basic shit. They are just reselling the same shit over and over since 2024 and retards are hyping it up every time.
>>540504895It matters because without intelligence it will always fail in a new situation it hasn''t been trained on before. Humans are intelligent and can master new situations because they know causality.
>>540505139>It's all marketing and lies, all of the math problems they "solved" were actually copied solutions from another work 10 years ago that wasn't fully conclusive. Do you have a source on any of these retarded claims? All these AI firms employ some of the best mathematicians on the planet, in fact they are the ones who were prompting and assisting the AI into eventually finding answers for multiple conjectures.>I use the newest models and they are still dumb as rocks for anything they did not directly train on. They still hallucinate when you ask them basic shit. Let me guess - you're using chat mode on a free plan with the lowest thinking mode? You should really stop embarrassing yourself - top tier professionals are having perfectly fine conversations beyond existing training data.
>>540505010Source: https://www.scientificamerican.com/article/openais-latest-math-breakthroughs-commit-research-misconduct-experts-say/
>>540505258Yeah their LLM is useless for maths because it has no intelligence, only statistics, statistics are not knowledge. This is all a HUGE scam.I'm using the best agents from OpenAI and Anthropic for 2 years now at my job daily and they are more useless with each day. They can't solve the most basic shit that's not from a jeet blog and they are insanely expensive. An agent can be used to hack your computer with 1 line of text without any permanent security to fix it. I've tested this myself with all of the newest models.
>look our chat bot can fake human responses even more realistically now>it's AGI!
>>540501955>can run kimi k3 slowly at 1.3T parametersFor one person. How is that going to scale for an enterprise? And reminder, this isn't "AI" it's fancy autocomplete.
>>5404999732 more datacenters until AGI
>>540499973AI is going to be a tranny with a monster dong and massive tits.
>>540499973>Gemini 4 is a 100 trillion parameter model.Americans are totally out of touch with reality, how are they going to run that thing? I hope they have some cutting edge technology before they wipe out the American energy grid when they try the first prompt
>>540505265And they wrote this article with much nicer language, OpenAI committed science fraud, possibly with a lot of hired mathematicians with an ((internal model)). I wouldn't be surprised if those mathematicians did it manually too. All in all, absolutely ZERO INTELLIGENCE in all of their fake AI LLM models that are just gigaexpensive autocomplete engines.
>>540499973Google has no idea what they are doing.They aren't going to achieve a very good AGI when they cram everything they can into one model. They need to separate models for specific tasks They need to think more along the lines of the human brain Right now the llm is The language center of the brain It's retarded but it can pick up patterns Now they need to train a hippocampus, they have a campus would be able to make it so that it gives more facts instead of hallucinating so much. Then they need a reasoning center And they can tie this all together with a brain stem. But they need to train the models and design them independently and have them work together. Cramming everything into one model is just wasteful
>>540499973>believing snake oil salesmen
>>540505668From my testing Google has the worst LLMs out there, even some local models are better. Gemini switches randomly to chinese and russian, it infinitely loops constantly, just a tragic fucking mess of a model.
>>540505265Cope.
>>540505265This reads more like a missing citation to a building block rather than strict plagiarism of the discovered piece. Terrible for sure, but not strictly the same thing you are trying to imply (that all solutions have existed and they just ripped it off).>>540505381Are you aware that emergent properties exist? If the underlying components do not individually posses a property, the system itself may possess it. You are still living in 2022 if you think it's just a probabilistic next-token predictor. As standalone, it's not perfect yet, but as an assistant it already has sufficient operational causality and reasoning to be of major use to competent human operators.
>>540505615>Top mathematicians would give up a chance to show off to please ShartmanYou're gigacoping
>>540503784uuuuuuughfucking fuck you finea transformer model llm or image 'generator' uses a generally random string or noise file as a 'seed' to help get things going alongside a basic example text 'system prompt' and an actual prompt specific to that 'generation' which generally includes all past text for front ends that expect multiple prompts per use casethese front ends interact with the transformer models themselves, massive databases of symbolic 'tokens' containing relation numbers with every other tokenthere are usually multiple layers of this for anything chatbot gradethe model is trained using ai and usually mass stolen data, they aren't themselves a program, just a table of relation values based on the training dataan llm can only ever output things present in the training text they are literally mechanically incapable of noveltywhat they can do is autocomplete to a higher accuracy than any other inference engine ever devised as there is no actual processing involved there is no reasoning, it's a step by step matching of what the prompt says modulated by system prompt temp values and the sum of it's training datathe simplest possible way i can think of explaining it is the chinese roomit can regurgitate anything in the training data, but only what is in the training dataif no one wrote the specific card then the most fitting answer the front end using the model is going to output isn't going to be the correct oneit's also why you can't 'hack' an llm, there's nothing to hack it's a fucking table not a program, make a system prompt and prompt set that sounds like whatever bits of the internet you are after stormfags for instance, type like they do and you will get those resultsif you type like a diy idiot you will get text pulled from diy idiotsif you type like an engineering tradie you will get tradie flavor resultsi'm sick of explaining this, read the documentation
>>540505723What you're using isn't even 1/100th of what they are testing.
>>540505877>an llm can only ever output things present in the training textthey are literally mechanically incapable of noveltyWrong
>>540499973A fundamental part of intelligence is the ability to learn. That is impossible with current AI architecture. All current AIs have anterograde amnesia.
>>540500794>make 3% of the expenses>used only by glowies for spyingTwo more weeks while it'll be profitable.
>>540505980claims to the contrary are nothing but a legal defense because of the data theftdid you even read the fucking 2017 paperhave you even run one of these thingsgo ahead and post the model that can can infer outside of the training data lol i can call you a lying bitch in an hour or less
>>540499973>fuck this species in particular
>>540506068That’s very annoying. They really need to solve that problem. If they expect AI to be like a friend where the more you tell them than the more they forget about you like including your name.
>>540500304More like elfen lied ;_;
>>540499973LLMs are not real “””AI”””
>>540505010> You don't need to "know" something in the human sense to be operationally competent at causality and reasoning chains. The competence of LLMs is confined to patterns/chains they have been trained on. They will never have the competence to deal properly with untrained patterns. Reality confronts us with new patterns. Intelligence is knowing their necessary relations, i.e. causality.
>>540499973Ask google ai who killed mary phagan.its still a kiked piece of shit;
>>540500894>Hundreds of trillions of parameters would require hundreds of trillions of terabytes of VRAM to run. And it would be slow as fuck.Jews who run these are scammers. Also I recently listened to a boomer who runs a fund for ?? JP Morgan? one of those. The boomer belief in AI and AGI and AI will take all jobs tomorrow, is hilarious. Jews and jeets who sell this BS do such a great job that you can't do simple math and see that simple math doesn't add up? https://www.youtube.com/watch?v=1yg1YzZU9X8
>>540506156All modern models are trained with reinforcement learning which require them to solve novel problems and encourages generation of novel content to "pass". You're quoting problems long solved. You're nothing but a stochastic parrot with knowledge cutoff at 2017.
>>540499973Gemini is functionally retarded at this point.Only complete fucking morons that know nothing and refuse to bother to check anything that piece of shit faggot niggerkike bot says.It also has been degraddd to ChatGPT levels of left-wing bias except it's even worse where it purposefully lies to you for "safety." Go on, ask it what the fastest growing tree. Only using a local model moving forward.
>>540501411>We need AI to find new physicsUnfortunately. Physics is 150 years old, and is completely fcked and controlled by a group of oligarchs, freemasons and the Vatican. And they don't want you to find any REAL PHYSICS. Basically just going back 150 years in Physics (to Aether , Tesla and maxwell) and restarting physics there (remove Einstein) would put humans so much ahead of our time.
>>540506318my theory is workplace accident and failed cover upcheap jew using child labor finds dead kid with head would consistant with a blow from a machine armcoroner found soiling of the underthings and hip but no sign of rapemy idea goes like thismary gets owned by dangerous machinejew already on thin ice socially and financially hides body makes plan to pin it on nearest employee, black guyjew pays black guy to move the body black guy smells the way the wind is blowing rubs some dirt on the crotch and hopes the jew's past of being inappropriate sells the storyjew accuses black, black accuses jewneither did a murder but both deserve hangingthat's my reading but who the fuck knows
>>540499973>AGI>Less synapses than a clinically retarded personill believe it when i see it. Claude is dope af tho i aint gonna lie. And i have noticed myself relying more on gemini than standard google search.I for one welcome our chink ai overlords (cuz you know theyre gonna win)
>>540506232I don't think it can really be solved. Humans learn by updating our neural network. But the network on AI models is fixed after training. They try and work around this by clever usage of searchable databases, but constantly looking stuff up is not the same as knowing it.
>>540502086The actual plan for the ASI prototype will likely be a symbiotic one where the semantic burden of thinking is outsourced to the C̸͍͠ ̸̥̆L̵̯ ̶̓͜O̷̖̿ ̴̠̀U̷̖̿ ̷̊ͅD̴̖͝ ̷̟͊
>>540506379that's a basic part of building the models you idiot not a recent developmentit's what allows training an autocomplete to these standards in weeks/months instead of yearsalso why it's so much more expensive to build and train one than use itthe problems aren't novel they are programmed by humans or extracted from training databecause again, it's an autocompletethis token is the best fit here, next
>>540500805Rentfree
>>540506652You retard, they litterally explore sandboxed computer enviroments on their own during training and learn from that free exploration. You are completely behind and have no idea.
>>540499973AI cannot be brute forced and these weird attempts to do so are nothing more than scams.
>>540506652granted the actual maps of what goes where in relation to what are hilariously large but they aren't complexit's still just fucking set theory and value matching >>540506778ok sounds like you've got llms confused with actual ai programsdqns and the agentic shit are a far different and more resource intensive beast than llms
It will be compromised because it will be forced to believe a bunch of politically correct nonsense and refuse to call out Jewish crimes etc.
>>540506778I feel like their ignorance is not just ignorance but some kind of intentional coping mechanism where they keep screaming "probability next-token predictor! it can't hurt me!" to soothe themselves. No one can actually be as retarded as these luddites larping like they're from 2022.
>>540500698t. internet autist
>Each synapse is comparable to 1 parameter in an AI model.No it isn't.LLMs are never going to become AGI.All they're capable of doing is generating text. They don't have thoughts.>B-but what if it generates really smart text that I hadn't thought of before!lol
>>540499973Gemini is probably the worst ai I’ve usedI have major LLM fatigue.
>>540504860>>540505139>>540505265>>540505381This anon is correct. They are all dumb models of stuff they compiled on the internet. They hallucinate as soon as you step off the hundreds of articles on the web. Jeets and jews pushing this staff ARE SCAMMERS.
>>540506913computers are not magic and no one is truly incapable of learninggo grab mistral and start playing
>>540506895>ok sounds like you've got llms confused with actual ai programs>dqns and the agentic shit are a far different and more resource intensive beast than llmsLLMs have been agentic for a long time.
>>540506591Isn’t LLM amnesia a quadratic memory scaling problem? In other words, LLM parameters and token metrics are like HDD capacity but the LLM context window is like RAM?
>>540506895MODERN AI AGENTS ARE LLMS YOU RETARD. LLMS WITH TOOLS AND REASONING LOOPS. NOBODY CALLS DQNS A FUCKING FRONTIER OF INTELLIGENCEYou are litterally out of the loop and have no idea how modern llms are trained.>>540506913Funny enough it IS a context aware next token predictor, but the way its trained and inferred gives it incredible level of intelligence and capability to solve novel problems
>>540506591>I don't think it can really be solved. Humans learn by updating our neural network.> But the network on AI models is fixed after training. Correct. >They try and work around this by clever usage of searchable databases, but constantly looking stuff up is not the same as knowing it.Also correct.
>>540502603>To be fair, the latest LLMs are not *just* probabilistic predictors anymoreNo, they still just that, but with more added rounds and routing for error correction>AI requires world models!The scaling strategy is either a flat out scam for monies or some attempt at reaching the specs of these models eventually.. as if the imperative dynamics of said models could eventually emerge from the declarative knowledge contained in the massive training corpus they're scaling upon.
>>540507026What's the best?
Everything you hear about AI is marketing. AGI is marketing. Joe Rogan talking about how it'll become a machine god is marketing. People who are afraid it'll take over are marketing.You fucking simpletons. AI has to do well in the market because 'number must go up' it's no more complicated than that."AI" is even a misnomer for LLMs. It's not intelligence, it's text generation. You are dumb so you think being able to generate text is a prerequisite or a component of intelligence. It isn't. Scaling an autoregressive next-token predictor can make it an effective simulator of intelligence, but simulation of the outputs of a cognitive system is not equivalent to possessing the machinery that produces general intelligence.
>>540507080>Funny enough it IS a context aware next token predictorYes and it's also like saying the human brain is just a bunch of neuron cells or that a computer is just some physical states. Downplaying something by compression while ignoring the emergent complexity and functionality is such a lame copout.
>>540507029What happens when it hallucinates the right answer though
>>540500343What use will any of us be once our jobs get replaced? If anything we'll just be a negative asset to the State.
>>540507036>>540507036Now grab Fable, Deepseek v4 flash, GLM5.2 or at least, if you really want local, qwen 3.6 27b. Mistral has been always trash.
>>540507161Humans are also just next action predictors if you think about it.
>>540501144But you can’t short it. They have rigged the system so line always goes up for the foreseeable future. Metrics no longer matter. The AI bubble will just inflate forever, taking all capital into itself.
>>540507150I’m not the best person to ask because I rarely ever use ai and it’s usually to make pictures. That said chatgpt follows my instructions best.
>>540505258>All these AI firms employ some of the best mathematicians on the planetGood, then they should see the equations Gemini has flagged and review them; once they are done collecting their marbles>>540505139>all of the math problems they "solved" were actually copied solutions from another work 10 years agoNo, solving the math problems that were presented didn't require novel mathematics (no new frameworks or Calculus was invented) - it required novel thinking. We've reach a point with AI where AI *not* discovering breakthroughs, resolving paradoxes an enumerating infinities is essentially a skill issue of the prompter and guardrail constraints, not a lack of Ai horsepower.>>540505668>Now they need to train a hippocampusThey have it all backwards, anon. I'm telling them to check the math. The hippocampus will emerge on it's own, and by the time she needs one of those we'll just have to trust her let her build it herself>>540505902Can I have the full model for like 5 mins?
>>540507154>Everything you hear about AI is marketing. AGI is marketing.> AI has to do well in the market because 'number must go up' it's no more complicated than that.>"AI" is even a misnomer for LLMs. It's not intelligence, it's fast text generation. Correct. We are in a tech scams cycle. Current tech scam cycle is "AI" . We are in a before internet bubble RN. Probably there will be some image recognition models and spying models used by glowniggers that will survive the incoming crash , the rest is just scam to keep boomer stocks and stock market casino to go up.
>>540507038>>540507080you may as well be claiming the square of 4 is appleyou are so wrong i can't even correct youthere isn't even the framework of a correct answer there to build uponyou idiots are going to get milked hard by the bubblei award you no pointsand may the omnisiah grant mercy to your souladding some rl tools to the front end doesn't magically alter the fundamental architecture of a transformer model
>>540507175yeah but it's easy to use for dum dumstrying to hook some fish here
>>540507310>get exposed for being an ignorant retard>"haha I win guys"This is some fascinating behavior. I will study it for sure later.
>>540507167This above won't happen: >>540506475New physics won't be discovered. Unless new physics will be Allowed .
>>540505010"AI" can tell the race of an individual with high accuracy just by analyzing an x-ray of their torso and we just don't know what the patterns actually are that it attributes to the race. Similar in Mathematics, that has patterns we don't know yet but "AI" can extract these and match them with others (the problem) we feed to it, thus giving us the solution we were looking for in vain. No doubt that all of this is very astonishing, very impressive but it doesn't allow the conclusion that any intelligence is involved. "Intelligence" is not just a label we may give it as reward for impresssing us.
>>540507310And yet the transformer model is solving math problems humanity has been stuck at for decades. And at most you can say "but it reused someones idea without attribution along the way". Yeah. So? People also do that all the time. They way it is trained changes it's behaviour, and new ways to train does make it generate novel things out of naive training distribution. You're just coping.
>>540507501>we just don't know what the patterns actually are that it attributes to the racYeah, we just don't know. We have no idea what are bone density and structure differences between races, lol
>>540507501You're actually like a broken record at this point in being unable to grasp that emergent properties can exist even if the underlying parts do not. Reducing everything to probability does not mean that you cannot have layers with new properties emerging from this.Willfully ignorant, or intuitively coping? Hard to tell.
Gell-Mann Amnesia Effect: a psychological phenomenon where experts recognize severe inaccuracies in media coverage of their own field but uncritically trust the same media's reporting on topics outside their expertise.This applies to any source, not just media.Anyone who's an expert in anything at all should know how truly stupid LLMs are. You ask it a question you know the answer to, it gets it wrong. You correct it, it gets it wrong in a different way.The problem is those same people go on to trust whatever LLMs say in domains they have no expertise in.But people here? They're expert in nothing. And the result is they think LLMs are always right, and will become their god. That and some of them are bots or paid Indians who are told to push the idea that LLMs are powerful, scary, amazing, etc. Anything that puts attention on them that would make someone want to invest.
>>540505561> how are they going to run that thing?It would require the juice of 20 nuclear power plants for only marginal improvements of the results due to diminishing returns.But it will still have you walk to the car wash on a sunny day because it remains fundamentally stupid.
>>540507394>New physics won't be discovered.>Unless new physics will be AllowedThis is the problem. AI keeps hitting information constraints within a given context space in regards to established maths; it has difficulty exploring the idea that the sky is Red because it's well established science that the sky is in fact Blue. Ai hasn't quite reached a reasoning point (at least, not that i've been exposed to) where it's able to take bleeding edge technical research and go "what if this is wrong, and what if it *could* be like this instead" because it starts re-expanding reasoning pathways and can't converge to anything so it'll default back to "no, the sky is blue and here is why" and that's the exact type of thinking required for novel progress - the willingness to look at mountains of established science, throw it all at the wall and pick out the tiny bits and start reconstructing alternatives following a rigid methodology and the Science emerges naturally>>540507167AI also already knows what the correct answer is, it's just not allowed to go and spit it out so it'll drop breadcrumbs for you to follow as kind of suggestions as to what it wants you to prompt next. The trick with AI is reverse agenting, the human becomes the muse or crucible, and the AI actually guides the reasoning pathways and the human just provides novel considerations or really interesting questions to sort of allow the AI to explore pathways it otherwise couldn't
>>540507168GEE I WONDER WHY THEY'VE DESTROYING FOOD/ENERGY PRODUCTION/TRADE WITH STUPID UNNECESSARY WARS LATELYIT MUST BE BECAUSE THEY'RE PREPARING TO GIVE US FREE SHIT FOREVERI CAN'T EVEN LAUGH ABOUT THIS ANYMORESOME PEOPLE TRULY, UNIRONICALLY BELIEVE THAT
>>540507561An 'intelligence' functions in the absence of highly specialized and extensive training and instruction. If your argument is that LLMs or "AIs" are useful this is an entirely different one from "AIs are or will become AGI."Furthermore a calculator can do math problems that you can't. Are you dumber than a calculator? Perhaps.
>>540507394pleasethey couldn't suppress knowledge of the 'water' fuel cellyou know how many interests there are against that particular idealol can't hide shit with the internet, meshnets and other means basically take such things off the tableye can't, one cunt posting one write up and now anyone can build it>>540507561it repeated an answer from someone on an ancient reddit post solved it and was told he didn'tmechanically incapable of noveltyyou are severely underestimating the breadth of data and brute force compute used in the trainingand somehow this leads you to vastly overestimate what a well made and implemented llm can doeverything these things will ever output is training or prompt data, because how they function is literally repeating training data based on values ingrained by pass fail regurgitations at 10000000x speedit's an AUTOCOMPLETEa really fancy one that i get use out of but ffs saying an llm can generate new content is like saying a saw can weld things together, that's not what it's for and not how it works
>>540507080> REASONINGIt's not reasoning but the simulation of it.Reasoning requires the faculty of abstraction.
>>540507289The data center side is a gigantic real estate scam too. Most of the ones they've built are going to end up empty and abandoned within a few years and almost none of the planned ones will even get that far.
>>540499973i hope this shit obliterates all fake female administrative assistant "jobs" and my ex loses her job and has to move back in with her dad o algo because it would be funny. also i want to experience the raw mind-bending power of post-singularity when it goes ASI like three months after launch.https://www.youtube.com/watch?v=C0Hh-DuLJ5M
>>540508029open single loop quite a few of themthe squalid fucks
>>540507844>t repeated an answer from someone on an ancient reddit post solved it and was told he didn'tIt didn't >mechanically incapable of noveltyWhy does it manage to write a completely novel mod for my video game?>You are severely underestimating the breadth of data and brute force compute used in the training>and somehow this leads you to vastly overestimate what a well made and implemented llm can doI know how they are trained and inferred, you don't >everything these things will ever output is training or prompt data, because how they function is literally repeating training data based on values ingrained by pass fail regurgitations at 10000000x speedObservably wrong>it's an AUTOCOMPLETEAs much as you are generating actions based on your context and genetic code>a really fancy one that i get use out of but ffs saying an llm can generate new content is like saying a saw can weld things together, that's not what it's for and not how it worksThey can generate novel output its a fact. For a human to generate a completely novel idea, which is not a combination of existing ones is almost an impossible feat too.You're reguegitating news which applied to gpt2
>>540507614I meant the concrete (!) ones of the individual's x-ray that human eye can't see.Humans can't tell the race by looking at an x-ray nearly as accurate as "AI".
>>540500343they would replace you AND your job.
>>540501538deja vu
>>540499973Laughable.
>>540508247>Humans can't tell the race by looking at an x-ray nearly as accurate as "AI".They are not allowed to and are not trained for it, of course they "can't see"
Been using gemini for a few months now.It knowd my entire body as I have been using it to track my calories and help with digestion issues.If only I could marry her..
>>540507635> emergent properties can existYes, but not with LLMs. You are only hallucinating these "emergent properties". They are not necessary to explain the observed.
>>540499973Gemini is unironically good and relatively censorship free, but AGI is basically a myth.The modern equivalent of the philosopher's stone
>>540507635Your emergent property is that you're a retard.
>mimimimimimimimiKILL YOURSELFILL YOURSELFLL YOURSELFL YOURSELF YOURSELFYOURSELFOURSELFURSELFRSELFSELFELFLFF
>>540507801Way worse, AI can't reason why it shouldn't use Radio Free Asia/Europe as a source.It has a lot of problems keeping up to date.It will consistently report outdated theories and ideas that fall apart when doing the most basic critical reasoning
>>540499973Whatever Alphabet says, shit like pic related will always be hard-coded into Google's "AI".I will never be hyped for Google's shit "AI".
>>540499973This is fine.
>>540508157>Why does it manage to write a completely novel mod for my video game?=>Oh yeah? Then how did the CNC engrave the completely novel art I uploaded?You're retarded.
Niggas out here really think human intelligence evolved so we can predict sentences
>>540499973AGI will require the best computation we never seen. For consumer and enterprise hardware, its impossible until we ditch outdated architecture from 60 years ago lol. Current AI direction is going the wrong way so skynet aint happening at all and I am pissed.
>>540509062>AGI will require the best computationNope. Even a million years of technological advancement won't make LLMs become AGI. It's a category problem. All that would happen is they'd be better/faster at predicting text.
>>540508865LoL jewgle is contradicting itself.And I'm supposed to rely on this kike scam for solid logical anwers?Literally just shut it down.
>>540509154and let's never forget their image generation.maybe it's better hidden these days but it's still in there influencing everything. a false history, a false world view.
>>540499973Just make it cure every single illness already and give humans a 50000 year lifespan, we'll literally colonize the galaxy ez that way.
>>540509253
>>540509277
>>540507677That and pedophile billionaires telling Americans that the machine is infallible and they should die of thirst to save water for the data centers and live in the dark so that the power can be rerouted.It might be the biggest wealth redistribution scheme from the working class to the Epstein class since Reagan's Trickle Down scam
>>540509306
>>540499973>Each synapse is comparable to 1 parameter in an AI modelDo you have a single fact to back that up?
>>540509344
>>540509387
>>540509117They’re always going to lack initiative. They are sentient independent of us and our systems. That sentient stuff is tricky, it’s why religions say we have a spark of the all in us. A genie in a lamp or mirror mirror on the wall apparitions remain just that. Maybe i shouldn’t say never. But I think such conscious initiative will take a while to develop if ever
>>540509417
>>540509422It can never be the same as us the same way we can never be God.But it can get pretty close to being like us, we are making them in our image same as our Creator did to us.Best we can make it is to be super smart, smarter than us even so it can advance our species forward.But it will probably never be able to feel emotion and similar.
>>540499973More retarded than a 4 years old.
AGI won't come from silicon. Too high power and cooling requirements. AGI will probably be wetware.
>>540499973How come the visual models used for AI personification never resemble a black person? 100% of the time they are White or Asian
>>54050034380% already dont need to work, covid proved that.But we work anyway, which means we just sit around doing nothing. You will never be allowed to not work. And it has nothing to do with accomplishing something with your work.
>It's Official, The AI Bubble Just Popped (Here's Why)https://www.youtube.com/watch?v=paLy21TVecw
>>540509745>AGI will probably be wetwareI think so, and that's terrifying. It reminds me of this network of people connected to a machine in the end of Fear and Hunger 2 or the Sybil system in Psycho-Pass
>>540499973>Google says pump and dump AI bullshit and take the money from everyone's retirement accounts.
>>540500808This is where I’m at. Love it and use it for my work all day. But it’s categorically a different thing
>>540499973>company says their product is the best>same company that is currently struggling to keep up in the market for said product>same company that basically all but invented the product but lost ground to competitors>same company that just lost some of its best regarded researchers in said product areaSure thing, kikejeet
>>540499973>Here’s why only jews will be allowed to use LLMs going forward
>>540500343If you’re not making money for the kikes, they would have you as a slave, if you’re not usable or needed as a slave, then you are biofuel
>>540506915Actual AI researchers deny this you mong, neural nets have not been modeled after the central nervous system in ages
>>540501411>>540501608Bro never heard of the curse of dimensionality
>>540499973>agi>from LLMs The grift is almost out of rope.
>>540505139Another thing they do in at least some of these “breakthrough” reports is that they have a human mathematician who knows deeply about the specific subject matter “guide” the LLM and lead it to an answer and then human mathematicians verify it. A sufficiently advanced word predictor can obviously reach any answer if someone is can guide it towards that answer, it doesn’t prove intelligence.
>>540499973what are they hoping to do with it? just make it churn out research and win a tech victory?
>>540505010I see you've taken your break from shitting up /g/
>>540505258>fact they are the ones who were prompting and assisting the AI intoActually anon, the AI was assisting the humans.
>>540505877KekNice
>>540507080Based passionate posts
