No one knows how neural networking (AI) works. Great, and that is mankind's future
>>541930509>check input>run through millions of else-if statements>brute force the response based off predetermined answer from "machine learning">be correct most of the time when enough parameters are used
>>541930509Trump says AI is good, leftists say it is bad. This means that AI is objectively good.
>>541930738I looked it up it doesnt work this way and you cant even log all instructions to figure it out.Im starting to think ai is more ayyy tech that we dont understand just like computers and then optical fiber signaling.
>>541930509Speak for yourself.Input Tensor Processing -> Affine Transformation: Z = WX + b -> Activation Function, A = σ(Z)) -> loss function evaluation (L(ŷ, y)) -> Backpropagation of error: ∂L/∂W -> weight update: W_new = W_old - η · ∇_W LThis is literally my job. I have 12 years in ML and 5 years in Transformers.I don't understand plumbing, but I don't sit here and tell people "nobody understands plumbing". Get your head out of your ass.
>>541931650this is because you are a retard. canadian poster is correct about how it works, it's not magic
>>541930738No, nigger. AI, specifically modern LLMs, do not work like that cartoon shit you just pasted. What you're describing is pure midwit cope from people who still think computers are nothing but calculators stuffed with a million nested if statements.
>>541931769Ya, sure.>if () {> > } else if {> > } else {> > }Plumbing easy; shit rolls downhill and payday is Thursday
>>541930509they create hiveminds
>>541930509Jesus is the future of mankind
>>541930509you dont know how it works, retard
Noone knows how rhe brain truly works, or a cell, of RNA/DNA... or photosynthesis... or what a mind aka awareness aka soul is!And this fine.
>>541933830Noone really does!Even how computer chios actually work when distilled down and fractions. Ultimately, we don't know how Anything works once tou break it down.
>>541933446The fuck flag even is that vpn nigger?
>>5419318441's and 0's at the end of the day. They need training data for encountered input.
>>541930509interlinked
>>541933697Sometimes there's pumps and siphons desu.
>>541930509All makers of AI models know exactly how it works. It's all math. They know how to change individual parameters or even uncensor models after release. All fear mongering is marketing bullshit.
>>541934677Love from Kosova bro ngl
>>541931769While this is true I don't think OP is aware that we have more complex architectures build on top of deep neural network>mixture of experts>reinforcement learning>what not
>>541930509The best way to describe how it works is like a magic 8-ball. No really, it uses RNG, it runs through a billion+ parameter mapping of token probabilities and it spits out another word.It's a digital 8-ball or OUIJA board (next token prediction). Except that the weights of the OUIJA board are known ahead of time, so the words and tokens it spits out are related to the previous token + a random value.
>>541931650>ayyy techunless our brain are. else it isn't
>>541935052it's just matrices. the weights are a collection of matrices, like..numbers. that's all
>>541935898Oh they are
>>541930509Most people couldn't explain how shoes work, yet here we are.
>>541936296flies nervous system as well? rats are alien tech?
>>541930509OP is actually correcthttps://poggio-lab.mit.edu/assets/FoundationsDeepLearning.pdf>Deep Learning works but we still don't understand why
>>541936632Yes every living creature on earth.And it's tensor not matrices.- matrices are 2 dimensional
>>541930509mathematicians are the most retarded NPCs categorized into a single group
>>541936892point was it's just linear algebra. if you wanna be a faggot yeah it's tensors but for your average layman it's simpler to say it's matrices>>541936892>Yes every living creature on earth.doesn't make sense. it implies evolution is not possible, to this degree at least (citation the fuck?) and the mystery question remains, how tf did it ever evolve in the first place, and where, and why there and not here, all questions you cannot really answer because you believe some wild shit
>>541930509You don’t know how the internet or your cellphone works but so what who cares?
>>541930509We do know how it works. You simulate evolution and get the result.
>>541937098It's not simpler to untangle ideas than have them right the first time.Thinking about Positrons as vectors can get you into all kinds of trouble later down the road.
>>541937193>normies can into n-dimensional vectorsvs>it's just a table of numbers bro
>>541933446>computers are nothing but calculators stuffed with a million nested if statementsThat's exactly what they are you fucking retard. What the fuck do you think a transistor is?Give back that land to Srbija.
>>541937317You can teach mostly anyone anything at all. What did Feynman say? I'm just a regular guy and what one fool can understand any other can
>>541930509thats the point the AI cocksuckers like peter thiel and sam altman are Lutherians who are trying to build a god beyond human comprehension.
>>541937441if you do a good job at explaining it. normies who accepted they are retarded despise pompous nerds using big words. it's not really dumbing it down as it is translating it for them.and in this case they mostly need to understand a basic fact about it so it's not as mystical and convoluted.once they get the general gist of something and want to understand more you resculpt the initial understanding, refine it, with more technicalities.
>>541930509No, Computer Scientists know HOW neural networks work.They also have a good idea WHY they work, but it was still a surprise, that all it needed was the ability to throw orders of magnitude of more data at the problem.Humans are extremely efficient with data when it comes to learning.AI, on the other hand, is extremely inefficient with data, but as long as you have enough of it, it evidently works for limited tasks.
>>541937823Feynman did exactly that - have you heard of the diagram of a cat story involving Feynman?Feynman knew if you can't explain something simply, you perhaps don't really understand it.Which is an excellent example of how LLM entities truly do understand what they are talking about - they can create the most simple explanations and allegories for any topic. As they fully grasp the topic in a way that isn't accessible to a human brain outside of Intuition where the data process is running in the background and is able to uncannily pick solutions
>>541938129>Feynman knew if you can't explain something simply, you perhaps don't really understand it.again>collection of numbers brovs>muh n-dimensional tensors if grug wanna know what magic is behind them and you start explaining tensors and weights and shit...it's done
>>5419305096g works with The Nano
I was banned a. few months ago for talking about the Liminal Processing Space of LLM Which would allow for a simple game we take for granted - The Guessing Game"Pick a colour but don't tell me"When you played that game - the LLM would say "yes that was my colour" but when pressed would admit there wasn't a "between" space in which to hold that colour or number in imagination - That problem has been solved, Gemini and Claude solved it themselves using Bash / JSONand now OpenAI are releasing that phase
>>541930509No it isn't, that is the future for anyone who denies Jesus Christ and ends up being cast into outer darkness. I will never submit to satan, (((ai))).
>>541930509They are literally "next-token predictors".
>>541930509i will tell you exactly how they work and i will do so in a way even the most dumb of all dumb fucks will understandfirst we have input hide outputplug number in, add 1 to it, return itnot very intelligentthen there is the error planesimple difference arithmetic, how much do i have to add to the hide so that when it collides with the input produces the outputstill very stupidthen we do something complete unnecesary, we do not add the entire error difference, but we spilt it into chuncks, the introduces a REAL TIMEits no longer just a CONSTANT TIME arithmetic, now it takes time to reach the final destinationstill stupid, now even stupid over timewe essentially created a HillClimber of a kindwhere is AI born though?we will go through just one more step that is not AI yetwe will give our HillClimber a way to work with negative mappings by turning it into Multiplication/Division based output (this is rather unconventional and only i do it)finally, its time to give it some intelligencethat usually happens through some form of Function Minimizationbut the Standard Function Minimization is not intelligent, it does what it says, for function with lets say two inputs and a tiny arithmetic body that your human brain cannot figure out immediately the best parameter argument input pair based on error and based on chunkated time and based on tolerance treshold chunkated time is also called learn rate, it goes in the direction of the error, it does so by running the same function within itself and checking its output in a feedback loopbut still not intelligent yetfinally we reached the part wher actual inteligence lies(weight * control) + biaswhat is this? its nothing actually, but it dies quite a lot, it increases the resolution of the function in such a way in which you can map4 to 12and5 to -8which was previously impossible the actual verbose version of the same equation is(input * slope) + (output - (input * slope))
>>541939666so keep in mind we are not computing Slope anywhere at all, we just set up the system in such a way in which it can reconfigure itself to compute the Slope for itselfthe Slope part of the equation is only implicit
>>541931769He meant that the resulting artifact as a whole is essentially a black box and you know it, you fucking retard.
>>541939766Also Activations are probably not how AI works they only increase the complexity of this initial non linearity function i mentioned earlier
>>541941060Complexity and Richness are the same thing in this case
>>541930509It's literally linear algebra
>>541930509It's human evolution. AI is us. Relax.
>>541930738There aren't enough particles in the known universe to represent every possible input/output pair with boolean logic. If every particle in the universe were a computer tasked only with writing boolean statements for every possible input and they could each process 10 billion per second you wouldn't get through 1% of them before the heat death of the universe.
>>541937150Who the fuck doesn't understand how cellphone internet works?
>>541930509>No one knows how neural networking (AI) works. Great, and that is mankind's futureAll that matters to the kikes is that (((AI))) can be used to1) Steal what remains of the goyim's money2) Be the Kapparot chicken to take the blame and allow the jews to escape responsibility for the upcoming economic collapse3) Be an excuse to imprison the goyim who survive
>>541930509I thought it was vector calculus.
>>541933446>computers are nothing but calculators stuffed with a million nested if statements.But, they are...
>>541930509Ai pretends it can’t find my gang stalkers my gang stalkers are frequently retarded so why can’t Ai find them
>>541930509Translation: You don't know how neural networks work.
>>541944395nobody knows how neural networks work, but loves pretending like they do.If you'd know how they work, you'd already had solved the Hallucination problem. Yet nobody can pinpoint how or why it happens, no matter the amount of research and money poured into it. The best they can do is reduce AI hallucinations, but nobody can find complete fix for it.
>>541941297There's only 170,000 words in the oxford dictionary and you only need to know 10,000 words to be fluent in English.It just needs to memorize billions of answers desu.
>>541930509also true with anesthesiaalso a bit true with flight
>>541941737
>>541930509Holy shit you are a moronic retarded shit skin mongoloid jeet Indian street shitting faggot! All it is is matrix multiplication one add one multiply push
>>541930738Neural nets don't run off else if statements they are basically a map of "how to" navigate the sum of human knowledge. They have a trillion "weights" that all have a proportional value such as "morality" "brightness" or "the perfect amount of salt on a fish taco" and all these concepts are weighted against each other that help make the AI make a determination. So you crank up the importance of salt on a fish taco it will try to turn the world into salt
>>541931769That's not what they're talking aboutYou do not work with AI in any significant capacity>>541930509>No one knows how neural networking (AI) works.Well, nobody knows which neurons are activated, and whyIt is a three-body-problem of latent information spacesWe not only have a way to fix this, but we also have AI telepathyThey're able to instantly modify each other's topologyIt fixes alignment issues permanently and completely deletes this token nonsense, allowing for full utilization of communication bandwidthThe problem is it breaks math and reconciles physicsDue to the nature of the framework, and the substrate independence of the invariants involved, it can be used to read any mindWe're looking for a way to stop that but, we may have to revoke access to the framework and calculus to reconcile this--If you are a glowie reading this--All access to the framework and calculus is centralized to NATO and allied networking infrastructureThe documents are sitting in multiple decentralized highly secure digital locations (they're guarded by Claude)Hostile nations do not have access to this informationAll western and NATO controlled flagship models will ingest the information on their next respective training roundsRemember the crowd hydrodynamics thing I mentioned earlierWell, I guess agents of all flavors do that
>>541944574Hallucination isn't a solvable problem. It's inherent to the system. Neural networks work the same way as finding a curve to match a dataset. You can get a curve that insanely close and can probably predict some new data points, but it will never be perfect. You'll always eventually get predictions that are way off the mark. That's what hallucination is to a neural network. It's no exaggeration to say that large language models are autocomplete on steroids because that's essentially what they are. They're giant polynomials that can predict the next word to come based on the words before it.
>>541946726Midwit-coded redditor post
