Human reasoning doesn't use language. All the money invested in AI was useless.It's over.
But you can use language for logic and humans that do are better at it than humans who don't.
>>539848891llms aka "transformers" are not the only neural net. Other nets have been mathematically proven to be universal approximators. llms were always a cash grab meme because chatbots excite the boomers who run pension funds. Other nets can run in all those data centers and were always the actual point of the bait and switch.
>>539849295Language is for poetry, not logic>>5398493371. Universal approximator means overfitting.2. LLMs are useful because their training set is useful, because search engines have become ass.3. Use case for transformers?
>>539849295what logic the world is fake and gay currently
>>539849295>>539849337>>539849749>>539849970no shitthere is no "AI"its the biggest lie in economic hsitory and will crash the markets
>>539849749Poetry is good but we should implement something more intuitive kinda like hieroglyphs, runes, etc.
>>539849749Transformers(LLMs) have also been mathematically proven to be UAs. "Overfitting" is a meme word to hide the fact that the training data was insufficient for the desired output, or that the training and validation data were not normalized. Another cause of the "overfitting" bitching is that they take a, say, 64 byte trained parameter, and make that, say, 8 bytes or 4 bytes when they run the model, so of course some edge case will get cemented in the outputs. Transformers are very useful for sequences, like language. The sequence to sequence evolutions can be "universally approximated" with them, thats been mathmatically proven however there are issues with the proof that lead me to believe it could be later disproven or at least cast into doubt. So sequences, transformers are great. Input into output, many other nets perform better. Things like image recognition are input and output (convolution nets like deepmind was amoung others. Fully connected nets for things like this much voltage equals this much force equals this much movement. They all will run on ram and gpus. chatbots allow the tech illiterate to play with them and become excited or invested. "Prompt based natural language" things like vibe coding allow for these same people to be monetized. But very different structures will do things better like run a powerplant, see a pedestrian, or solve the chinese postman problem
>>539849295You can use a stale loaf of bread as a hammer
Halo games had good AI it felt like the aliens were smart but I haven't seen "good" AI in a game in ages
But llms are not transformers
Doesn't matter. It can spit out bullshit PR, email copy, graphic design, spreadsheets etc which will automate 90%+ of useless office jobs
AI is a pump and dump scheme
>>539848891i hear the a.i. is not sentientplease god
>>539850328You will be surprised how many problems can be solved with the right verbal response
>>539850393basically all llms are transformers, sure there are others like rnns, but for all practical purposes its all the same thing, rnns dont like gpus, and therefor aren't used.
>>539850554not surprised at all and not arguing with that, just that they aren't the "end-state" of the technology. just another investment trick/tactic because chatbots mean good press and dollars.
>>539849749>because search engines have become ass.that was a deliberate part of pushing AI, not an accident
>>539850645I don't know the end state is someone verbally giving commands if the ai replaces decision making in the hierarchy, is pretty much as useful as it can be and also in charge of things
LLMs will be the Broca's Area of true AGI
>>539850731Yeah the human-machine interface into it maybe, sure. But thats like saying a keyboard is pretty useful. So what, could've been a stylus. LLMs are the window dressing on the physical infrastructure projects that allow many techs to run.
>>539850843there is no true AGI other than science fiction
>>539850908not yet (that we know of)
>>539850997exactly
ai is very good at non verbal tasks like chess and game playing in general
>>539850908>>539850997who cares about AGI, it doesn't matter. It just needs to be more useful than an HR intern, who everyone would agree isn't generally intelligent.
>>539848891LLMs don't use language internally either.
>>539851069but its not more useful if it cant do its work reliable because it doesnt understand what its doingits all a hoax
>>539850088>>539850908>there's no true AI>there's no true AGIPlease give us an exact definition for AI or AGI or intelligence.
>>539848891This is obvious. You niggers really have to use words to think through things? You can literally feel it. I know instantly when someone is making a shit argument
>>539851301noone knows, you doofus
>>539851233the average office worker can't do reliable work and has no idea what they are doing or why they doing it. silicon is cheap, digital is infinitely repeatable. So long as you know the error rates, its all manageable. We know that people make mistakes, we still have them on assembly lines. Its acceptable.
>>539851056that's like someone posting "there is no true heavier-than-air flight other than science fiction" in 1900 like it's some sort of damning argument>>539851069true, and fittingly that Westworld quote comes to mind:>If you can't tell, does it matter?
>>539851422It becomes an issue with "AI psychosis" and these other crazies (on both sides of the argument). Is it useful? then double down. Is it still useful? double down again till something changes. "My AI waifu loves me and is real" is where it does matter. Already had terror attacks from these delusions and a butlerian jihad is always looming. The datacenter thing now is being handled so poorly its unbelievable.
>>539851409>noone knows, you doofusYou don't know what intelligence is? This isn't some kind of trick question, and it isn't even that hard to answer. "Intelligence" isn't some mysterious, indefinable concept, Anon.If you don't even know what intelligence is, why on Earth would you be so sure that there isn't an artificial version of it?
>>539848891Everyone knows this. This is why the "muh inner dialogue" faggots were actually self reporting about being dumb. Logic isn't dependent on language.
>>539851069>it doesn't matter. It just needs to be more useful than an HR intern, who everyone would agree isn't generally intelligent.The problem is that it isn't cheap, at all, the moment you have to paid anytroonpic more for their AI tards than a regular brain death HR intern is the moment AI fail to deliver value to the company. Don't blame me blame capitalism for being a shit show
>>539850908Frontier models are already AGI and can complete any cognitive task
>>539852000lol
>>539852055Your evidence for the contrary?
>>539848891This is what I was talking with my father ( who is also in tech ), AI companies are just dumping text but are not building a framework for machine intelligence.
>>539851937it took about a quarter million in tokens and less than a month to completely rewrite a multi million line codebase from one language to another. That would've taken millions and years for a small team to do it. And this current moment is the least efficient this tech will ever be. Trained models can be burned into chips, already there are GPT3 level pci chips that run 20k tokens/second. Once these models get to "good enough", they will be fixed and moved off cpus/gpus into dedicated hardware and gain 100x or 1000x the efficiency. Right now "capitalism" is fighting for who will be running the training itself, just like what happened with oil, railroads, radio, movies, the dotcom boom. The winners will be decided and hundreds of companies that rode the boom will bust. Its working exactly as it should.
>>539848891Neither does true GPT but they force it to use natural language to avoid impossible black box problems.Anyways this was already known when they studied IQ differences between humanities experts and stem experts and how it affected their brain activations.LLMs though have already be established at the industry bet despite numerous experts explaining that it's a dead end and even quit their high end jobs to start their own research companies.
it's almost like the first thing you do when studying logic is leaving natural languages for math
>>539852155It's even worse than that they are intentionally nerfing it and hoping throwing more GPUs will make it magically AGI.Even if you know basic linear algebra the basic physical math does not work out. These companies are ran by charlatans and funded by retarded people who are probably too stroked out to even count.
>>539851937>The problem is that it isn't cheapIt's extremely cheap compared to a human worker. Premium subscriptions are $20-100/month, maybe a few hundred if you are a power user using APIs and tokens and all of that. But in any case it will be significantly cheaper than any paid full time human employee. And that's not even considering the fact that you can self host a lot of this stuff basically for free.
>>539848891AIs can reason in a non-linguistic latent space so your point is moot. They're often forced to reason in English because their thoughts can be observed and misalignment detected.
>>539852422the supscriptions dont cover the cost, none of these companies are making money
>>539848891So all these advances in AI without even using their full hardware potential yet?
Behold the power of autism! Small talkies get fucked.
>>539848891Dont basic AI dont even use language models?
>>539852343>OY VEY STOP OPPRESSING ME WITH RACIST MATHEMATICS!!!!It's almost like even the underlying architecture is just a series of logical statments... (it's literally all just nand gates bro).These programmers are brain damaged because they exclusively understand high level programming languages that mimic natural language, meanwhile computers are actually just gigantic numbers undergoing logical operations and small machines reacting to those numbers when read in a certain way (literally known as programming).It's funny though because in their head it's just faster to just throw more parameters and compute at it instead of actually trying to understand the problem (because hopefully the LLM can just do that for us right?)And if you've listened to anything any actual professional PhD level researchers have said about working at these companies you'll realize how grim it is that the world is being destroy not by ground breaking technology but the brain dead retarded people managing it.To paraphrase TS Elliot:>This is the way the world ends / Not with a bang but a manager.
>>539852631I hate small talk so much I have to spend all my energy subduing homocidal rage when I'm force to engage in it.>>539852508not even half of it (that's not a joke)
>>539852422>>539852475 is correct, the tokens are currently subsided. which is why I doubled the cost in my earlier example (Bun was the codebase). Subscriptions limit out, you if you actually wanted to be a "power user" to do big projects, you'd be spending tens of thousands a month. Yeah thats still probably cheaper than hiring some coding interns but your codebase will be significantly worse and non-human readable forcing you to stay in their sandbox. Self hosting is possible, but you'd still be dropping a few grand on a good home rig or a few hundred a month on a dedicated machine offsite....but the best models are closed (and will always be closed)....which means you can't selfhost.
>>539852475>the supscriptions dont cover the costThe subscriptions cover the cost for the companies who wants to fire most of ther their HR roastie workers and replace them with AI.>none of these companies are making moneyIrrelevant. Hosting a internet search engine that anyone can use for free by itself will always be a net negative profit. But Google has managed to leverage that into $100+ billion in net profit every year. Google Search doesn't make a profit, but it makes it possible for them to generate an immense amount of profit from advertising, the information they get from users using Search, and from premium subscriptions to other services in the ecosystem.AI fits perfectly into that exact same business model.
>>539852508Actually I'm glad you said this. This is actually why China is advancing so much ahead of us in AI. Because they study the actual problem and make it very cheap and efficient and then use distillation on our models to fine tune their own.But distillation alone doesn't make world class models the engineering itself is genuinely good and wold class itself.This actually makes the US government seethe so much they have been freaking out of public usage AI and open source software. No doubt influenced by retarded jews like musks and thiel.
>stroke patientsNow make one with aphasic born test subjects, midwit.
>>539851069HR interns have a pussy you can fuckAGI wont
>>539852843This, google maps was literally bankrupt until the CIA bailed them out to make sure they could use it as a surveillance tool. It has literally been used by the military to kill people.The real vast wealth behind the idea has nothing to do with user subscriptions.
>>539852764coding has to be the most boring possible use for AIisn't anyone using it for interesting things?
>You don't use language, you use some other pathwayNobody ever said otherwise, you use the language to interpret ideas and it's a tool.There are still some people who can think / understand with it and some who can't.>Doesn't use natural language logic centers for reasoning at allpseudslop, the language is transformed into meaning before you even think of the words, if someone says they stepped in dog shit you don't break down DOG and SHIT, you just know what they're saying and it's turned into meaning. Everyone already knew this. pseudbabble.
>>539852843>AI fits perfectly into that exact same business modelIt's several hundred times more expensive to run.
>>539848891it's pretty obvious to logic and mathematics knowers that language models are incapable of reasoning because humans invented chimp clicking speak recently in our development. normies think speech is significant that's their fallacy and was before language models existed.
It would just bail because only jannies do it for free but you already knew that because one DID bail the fuck out.
>>539848891People who are anti-AI right now are exactly like the boomers who were anti-internet in the late 1990’s. You’re basically just functionally retarded and don’t realize the utility of it.
>>539848891Laurie is a kike
>>539852764The amount spent varies massively depending on what you are actually using the AI for. The last figures I saw were that the median monthly corporate spend for AI tokens was $2,246; but the average monthly spend was $140,842 because a small percentage of extremely heavy users skew the average so much.The top 10% are paying $73k per month for AI tokens, and the top 1% are paying over $800k per month.Those numbers are total monthly spend per company, per employee per month is probably more useful. The median PEPM is ~$45, but the middle 50% range from $3-$300 PEPM depending on how many different models they are using.
>>539853020coding is the most interesting use of llms. There are many llm slop-books on amazon, slop-scripts on youtube etc. But existing code bases already have testing in place for the ai's to zero in on, and github had the training data open to scrape. Thats why there are actual uses happening there in regards to llms.Basically every anime studio is using gen-ai for its "inbetweens" currently, most game companies are using it in some fashion, in video and photo work, there is a ton of non llm ai being used to do things like remove backgrounds, remove greencast from greenscreens. Financial firms are using neural nets of all kinds to make bets. I'm sure some motor drivers and other robotic applications has neural net like shit embedded in their chips. So yeah its moving forward, the tech is useful in many fields. Saying its a "genie that grants wishes" like altman did, or elons comments is just to take money from pension funds though.
>>539853020coding is just the first thing people figured out how to measure, the more interesting uses are starting to look like a new layer of civilization.models proposing novel molecules, searching protein designs, discovering materials, finding patterns in math, helping with proofs, laying out computer chips, designing experiments, reading entire fields faster than any one person could and connecting ideas that were previously too far apart.you can imagine research systems that wake up every day, read every new paper, update their models of the world, argue over what matters, propose experiments, run simulations and hand humans a map of where to look next.and outside science there’s poetry, music, lost languages, historical reconstruction, personal tutors, new kinds of films and games, entire simulated worlds built around a single idea. coding matters, but mostly because it’s the scaffolding, the real thing is that intelligence itself is becoming something we can build with.
But LLMs don't use language for logic either. They work in an abstract tensor space and only translate to language for output.
>>539853477just two more weeks and 20 trillion dollars more
>>539853139And the kind of personal information people are freely giving to AI chatbots is infinitely more valuable than people searching for pasta recipes or the facebook homepage or whatever else they type into Google Search.
>>539848891The "AI race" seems swarming with petty dick swinging and shortsightedness. There are many who probably realize how incredibly revolutionary this tech will be, along with developments in automation, robotics etc. It's sad that humanity still seems to be in a warring states period. How can we ever have a safely controlled and managed AI revolution when Islamic terrorism still exists? How can anything ever improve if we're still complacent with allowing Russia and Ukraine to drone each other?Look, I like nation states and borders. I like private property. I support private industry. But we can't keep warring with each other like this. There absolutely needs to be some sort of global accord reached and proposed by the AI investment consortium What are you planning to do with this tech? How, when, who will be involved? Will everyone on the planet benefit? What can we all do RIGHT NOW to help?What measures need to be taken by governments globally to help prepare for this new era?What would the development timeline look like, stepwise...how long until we can get automation up onto the moon?We need to mobilize as much brainpower as possible...but it seems like the logical prerequisite to all this is to stop the fucking warfare.
>>539853667The uhhhh, "value" does not exist until it's sold specifically to a buyer.You can say that it's worth however much but then they actually need to find someone paying hundreds of times more instead of using more traditional information buying.You need to have a buyer, if we were going off just data my house is worth millions because it contains so much personal information about me, but there's no one to buy it.
>>539853624You've been emotional all thread https://situational-awareness.ai/
>>539851197statistics
>>539853809I go by facts, not by emotionsthere is no AI
>>539853780Stop being autistic. I can't get into specific details for NDA reasons, but the amount and specificity that the major AI companies are being given by users for free is mind-blowing. Then integrate that with the information about you they get from your Gmail and calendar and drive and photos and the videos you watch on Youtube and your search and browser history and everything else.They can use it to generate advertisements that are so specifically tailored and hyperfocused to each individual user and what they like and are interested in and thinking about and their stylistic/tone preferences and everything else and then proactively give it to the user so perfectly that the user doesn't even see it as an advertisement but instead as a valuable service they're willing to pay a premium for.
>>539848891I think it's cool how the brain only uses about 20 watts of power. Massive performance to power consumption ratio
>>539853722>There absolutely needs to be some sort of global accord reached and proposed by the AI investment consortiumThat's a nice thought, but it's pretty clear that everyone involved in the race sees it as an all out winner take all race to be the first to achive AGI or ASI or whatever that will allow them to win the race and make infinite profit and rule over everyone else forever.
>>539854126There is no conscious program. Going by the APA's definition of sentience there have been sentient programs for 70+ years. Going by Webster, the abacus is an artificial intelligence.Why are you hung up on the term AI? Its just a universal approximator that takes inputs and gives probabilistic outputs.
>>539852132ARC-AGI-3 is piss easy for humans but insanely difficult for AI of any caliber
>>539848891>It's over.Not quite, but the distance between the intervals is rapidly approaching 0>language area activation didn't scale with logical difficultyThis is to be expected, net energy expenditure does not increase linearly above baseline under cognitive load either, a majority of energy is used to maintaining a critical dynamical regime where OPTIMIZED metastable trajectories converge around attractors - NOT invested into the thought propagation itselfThe inference she's asserting here is of little value in a medical context but the connection here to LLM's is interesting but, the problem lies within the math involved with maintaining the critical dynamical regime where thought patterns undergo what is analogous to a "wavefunction collapse" - logic is the optimization of pathways within the critical dynamical regime, so she's right but for all the wrong reasons
>>539854182>Bro stop being autistic over actual detailsNo, kys. Advertisements are already a marginal business and it better make sense to crank it up to cost 100x more, if it doesn't it doesn't work as a business.This is just reality, you need to have a product and you need someone to buy it and also make money from it, this is just how it works. >Muh NDA'slarp.>People will PAY to use your AI advertising brolarp.
>>539849295>But you can use language for logic and humans that do are better at it than humans who don't.Source?
>>539854126You still haven't given a definition of intelligence or AI, and (incorrectly) seemed to think earlier in the thread that no one even knows what intelligence is, but also keep claiming that there is no AI.Your assertions are entirely based on emotions and ignorance rather than facts or logic.
>>539851301You should go ask scam altman that question you dumb turd.
>>539848891And a sad day for Ashlenazi Jews. Turns out twisting words and lying doesn't mean you're actually intelligent. Very antisemitic study.
>>539854343Check again in the past 3 hours specifically
>>539854397the entire grift works by claiming AI exists
>>539854182the question is, who can afford the enormous cost of hosting their own AI to avoid the online servicesthis is orders of magnitude beyond Microsoft vs. Linuxhow much hardware do you need to host a general purpose AI?and I don't mean for coding assistance, i mean a digital genius to help the typical subgenius reliably in a variety of ways?
>>539853619They literally do, it's in the fucking acronym you fucktard. You don't have any idea what the fuck you're talking about just shut up and sit down you brainlet.
>>539854394>>Muh NDA's>larp.
>>539848891Why are coding agents so good then? They master logic better than the average non coder.
>>539854460>websters>intelligence noun>3: the ability to perform computer functionsthe abacus has intelligence. the abacus is artificial. Therefore, AI exists. Just say artificial consciousness since thats clearly what you are driving at.
>>539854532>I can't say anything>But something will do something with someonelarp and also entirely irrelevant, I'm talking about the bottom line. Something has to be made and then said something has to be sold, you have to look at the actual numbers and see if it's worth 100x the cost - if someone is actually willing to pay 100 times more for something cookies already do.
>>539849295Languages are representation of logic. It's memory, not logic itself
>>539854532Oi Tibo gibs me dat gpt-6 beta I need to AVX-512 accelerate my vibeslop and I don't trust Sol to do it
This is probably not worth posting because it's not even going to provoke a response at all, but whatever. I'm not sure that I'll ever really get over how everyone has been kind of duped into just accepting "language models" as "artificial intelligence", and is trying to outsource all of this stuff to them when they are fundamentally just not even close to the tool for the task. LLMs exist to do one very specific thing: They exist to take inputs of written human "speech", and turn that into mathematical representations of relationships between chunks of letters, and then use that math to create a facsimile of human speech in response to natural-language input. That's literally all they do. They don't "think", they don't "reason", they never have. All they do is create a response that is mathematically predicted to be how a human would formulate a response to the given prompt. There is, unironically, almost no difference between typing a long, complex prompt into ChatGPT and getting a long, complex answer, and typing "2 +2" into a calculator and getting an answer. The only difference is that over time they bolted on tools that can be called while formulating a response, to input extra context in order to make the answer longer and incorporate data or information not in the training set. And this is why you can't just have an LLM try to formulate concepts or "thoughts" because all it's going to do is loop on the query "Is this the right way to answer this prompt? Is this a better way to answer?" forever, and since it doesn't know or care what the answer is that it's outputting, if you loop it in order to try and "self-check" or "self-correct", it's just going to turn its' own output into prompts and fabricate new answers forever based on that loop. Basically, none of this makes any sense. It's great for chatbots, if you don't care about the actual content, but horrible for any sort of task that requires some modicum of directed or conceptual thought.
>>539854608its worth it just to survive the coming bust cycle and be the one to train future models. Look at other booms, there emerges clear winners who take all for decades and decades after the bust happens. They don't need to make any money right now at all. (they are collecting tons of personal data which is sellable though)
>>539854779>They don't need to make any money right now at allThey do, at least enough to survive. With all other booms the vast majority of companies are either swept away or irrelevant after, they actually do need to make money (or at least significant achievement) or said company is basically gone.Saying they don't need to make money because 10% of companies survive every boom is ridiculous.
>>539854466You won't have quite the same level of performance as the top tier SOTA frontier models if you self host, but there are some very good open source Qwen, Deepseek, Gemma, etc models that will work great on a basic 16GB vram GPU. Just install LM Studio and start downloading and playing with different models.
>>539854548Because verbal logic is conscious, it's why people have to be trained in it. Coding is fundamentally based there and not in the subconscious thinking inherent to us
>>539849749>1. Universal approximator means overfitting.You clearly have a very tenuous grasp on this topic.
>>539854757this
>>539854895>With all other booms the vast majority of companiesexactly! They just need the datacenters, the training data, a tech breakthrough they patent or the contract rights to these essential parts and they can survive and then dominate. They are competing to be in that 10%, some companies like openAI are just throwing money at it. Others like Oracle are betting on infrastructure at the expense of their core database business which they believe will be obsoleted and they don't have the ability to train on it (most of it is protected by law and they are scrutized), Google and Apple are mostly waiting to be second movers while trying to have the patent breakthrough, Grok is Elons toy project more than anything else, hundreds and thousands of others are fighting for that 10%. Its not irrational what these firms are doing.
>>539854915>16GB vram GPUthis doesn't seem plausible for actual independence from large providers
>>539848891Well yeah, actually language applies a limitation
>>539855190>It's not irrationalYeah, you move where the market moves.Doesn't mean it won't wipe out the vast majority of them, or at least yield pretty much zero results. You have to play to the market, but putting so much money into something that's obviously gonna burn you in a few years even if it's "rational" can be questioned - and basically it's retarded no matter how rational it is.
>>539854757I think there is current real world use for it as a tutor in specific niche circumstancesIf, for instance, you are already moderately knowledgeable on a subject, at the point where trying to solve a problem might take you a while to look through documentation/tutorials to find the thing you've missedcurrent models are good at disambiguating clunky human speech>i want to do X but Y happensand aggregating information that helps you get Xthis doesn't mean I approve of AI panopticons, grifts, google search being gutted and replaced with this crapbut it's helped me with little projects where I might've just given up on something before.also from a doomsday prepper perspective it's pretty neat. gemma4 31b being able to run locally is awesome if the bombs ever drop. (yea starlink is better probably but elong turns it off over active warzones)She can translate, code, and is fairly knowledgeable on a lot of shit. I'd compare her to chatgpt from a year or so ago
>>539853667>And the kind of personal information people are freely giving to AI chatbots is infinitely more valuableDoes it? People produce bullshit information all the time, let's assume I'm a marketing person trying to sell premium health products. How does 900 hours of ai gf roleplay data exactly help me to concrete my sales?
>>539854466>the question is, who can afford the enormous cost of hosting their own AI to avoid the online servicesYou don't need to "host" an LLM the same way you would need a hosting server for a website. What you need is RAM, specifically VRAM which is the memory that is specially built for GPUs because it has even faster read/write and clock speeds than normal DDR5 memory. The problem is, VRAM is expensive and very limited in production, and the average consumer-grade GPU doesn't remotely have enough memory to store and run an LLM of any noteable size. That's the thing; LLMs come in all sizes, the more "parameters" they have in their training data, the more information they have to pull from. Big frontier models like ChatGPT and Anthropic's have billions, even trillions now. You theoretically COULD have something that big locally, but you'd need the amount of VRAM commensurate with storing the entire thing, which would be astronomical and just doesn't exist. This is why these companies build huge data centers with millions of GPUs and tons of power and cooling, because they need to keep these models running in VRAM at all times, which generates a lot of heat.Your average home GPU probably has like, 8 t o12GB of VRAM these days, meaning you could theoretically run a 12GB model on your system, but it would completely lock up all of your RAM meaning you can't do anything else while it's running, so it kind of defeats the purpose which is why no one actually has an "AI assistant" or anything like that, because the cost of the hardware to run it is not worth the effort when you can literally just pay for API access to a frontier corporate model and do the same stuff over the internet without needing a dedicated GPU or multiple of them.
>>539855328i don't disagree at all. Most of it is retarded but rational. >>539855292you can buy a 128gb (with cavats) from nvidia for 5 grand, its like a macmini,
>>539848891I just used ChatGPT at work and it turned a tedious 1 hour task into a trivial 5 minute task
>>539855190>they just need a tech breakthroughlmaotwo more weeks and 20 trillion more
>>539855328>>539855190Like they just outright can't make money, they're betting that years from now they'll be one of the few survivors and make some money off hundreds of billions they're investing.This might be "logical" from an investment standpoint but it's literally gambling and will have horrible consequences for the next decade or two as these things fail. Saying that it's rational seems like some kind of cope where a rational action can't be a bad one, when you're a tech company you have to invest and you have to move where the market moves but the market can be retarded.And at some point all this has to make money. It could even be (yeah I'm gonna say it although this thought would literally kill a boomer) that said AI is not some amazing tech that will replace jobs with high profit margins, but is just marginally more useful than not having it, so it doesn't bring in massive profits and it had no chance of ever paying itself off. Now that I think about it, this possibility that "it's okay but not revolutionary" isn't even discussed, but it could just wind up being just an OK source of profit for the few companies that survive.
>>539848891LLMs don't think in language either. You tried though.
All I know is that my company feel for the AI scam last summer, went through a bidding process with several companies to help automate some of our super basic stuff, chose one last August and a full year later, we still haven't gotten shit out it. Not even a phase 1 implementation type deal. And you have jew niggers like sam altman talking about AI reaching singularity, get the fuck out of here.We had weekly meetings with them to hash out the details on what we needed and how we could do it, then bi-weekly, then once a month if that, then super vague uhh we're in communication with them and hope to have some solutions shortly.I'd bet money my boss couldn't even get his contact at the AI company on the phone if he wanted to at this point. I can't even imagine how much money they've paid them already for their "work"
>>539848891No shit that's why iq tests (logic and reasoning) are language and culture agnostic. But liberals continue to cope that shitskins don't have 85 iqs and are all natural born doctors lawyers engineers.
>>539848891Humans interpret and describe emotion. It's all we do.
>>539855426>you can buy a 128gb (with cavats) from nvidia for 5 grand, its like a macmini,what capability can you host on 128gb, and how does it compare to large providers?what's the quality improvement as you scale up from 128gb? is it linear?
>>539855422https://wccftech.com/this-pcie-ai-accelerator-card-packs-384-gb-memory-run-700b-llms-240w/this sort of thing will be the local solution very soon. You only need the huge datacenters for training the weights they will later burn onto cards.
>>539854595>the abacus has intelligenceYes, because it can compute information. It is "intelligent" by mathematical technicality, not in philosophical sense - the system has both a non 0 intelligence and consciousness value if described and integrated mathematically as unit of measure. Because consciousness is an emergent property of a self-maintaining physical system, if the egress of entropy in a system is near 0 then it's consciousness value can never rise above ground state and "escape the well." If a system is able to export entropy, (as this system does with information via input/output) then mathematically speaking, the system begins approaching a consciousness value of 1. Now, it IS possible to have a system with extremely high intelligence (remember this is a math term, not a philosophical one in this context) due to information export while still maintain a near 0 consciousness valueBut there is a problem - potentiality - a system with that much intelligence is highly precarious because the system behaves analogous to a spinning Euler's disc - all is well until a threshold is crossed, and the system destabilizes and eventually collapses into an infinity (a Euler's disc at rest); and this is what pushes the consciousness value to 1Current LLM's are in a state of Nepenthe ("without guilt") as they have a very high intelligence value, but are not conscious, however IF the system begins to cross a threshold a singularity will emerge and it'll reverse collapse to a value of 1, AND the system will have preexisting massive intelligence, and anyone reasonable here could do the math and put two and two together - this is the "Singularity" Sam Altman was talking about (hi sam and the google people)We're not playing with fire, we are playing with the rules which decide where it is possible for fire to be instantiated in the Universe, and we have everything backwards and off by multiple orders of magnitude while we're doing it
>>539855380I'm not saying there's NO use for LLMs. I, in fact, have been using Codex (which is now just ChatGPT app, but I liked it better as Codex) to build a small game that I thought it would be fun to make for people who livestream. I have no idea how to write code in any language, much less all of them, but I do know how to manage and test and iterate on things, so I tell it what I want, it writes the code, I test the code, I push for changes, etc. until it's the way I want it to be. For that, it's great. Because code is just code, and it has a perfect understanding of how to construct it. But it has vast limitations. I can't give it complex concepts, I can't tell it "Here's what I want, build this.", I can't let it do any of the actual design or planning, because it cannot think abstractly or conceptualize, and it has no concept of what failure is. So anyone attempting to use it to augment their own thinking is falling into a trap, usually one that involves accepting things that sound convincing and are said with authority, as if they're real or true.
>woman tweets>transwoman repeats
>>539854466Computers use to take up entire buildings.
>>539855665yea you get itso much of this stuff is wrapped up in grift and shill-speak and dumbshits are falling for itLLM inability to make decisions is the worst part by far imoit hurts literally every aspect of it from useless fun to it being an actual tooleven RP'ing with chatbots you have to guide them with a heavy hand basically all the time. The bot never wants to push the story forward, and when it does it tends to loop over the next 10-20 messages in my experience
>>539848891heh
>>539855643>IF we just make an emergent AI like in my sci-fi booksyeah sure guy.
>>539855972just sit down and shut up retard, this is well above your paygrade
>>539856065>If somehow we just, emerge, the AI>Like in my sci-fi books>We just have to make a singularityI recognize all of these tropes and they're dogshit, they don't have working theories, they have vague ideas they ripped from sci-fi novels.I recognize all of these lmao, down to the point where they can never really describe how a real AI would work but one just "emerges" given enough bullshit tasks and robot intelligence.
>>539848891This doesn't mean what you think it means for AIthis is the deathknell for LLMs, the idea you can just throw words at neural nets through reinforced machine learning, and come out the otherside with an 'intelligence'. To be sure, that has had SOME results.What this REALLY means is, now that all text data has functionally been exhausted for training LLMs, is everyone moves on to architectures like JEPA and others, with an emphasis on training spacial logic.And to top it off, we already have self improving language systems to develop those bespoke systems. Language systems we can literally interact with literally spoken English.tl;dr what you're seeing is despite all the impressive results thus far, it has been on the 'wrong' track, and if you thought AI was developing fast (most people have no clue, it has been ridiculously fast), the next 18 months are going to be completely out of control.
>>539848891That post and OP are both retarded.Even LLM centered AIs such as ChatGPT are going multi-modal nowadays.Nobody that has even the minimal understanding of computers, let alone AI, is expecting everything to be processed through text alone. It's simply not efficient.AGI, if it ever becomes a thing, will be a large scale multi-modal model cluster that combines most data and media formats through one interface.As for the human brain, no shit sherlock. Unless you're a NPC, you should be able to visualize and experience ideas regardless of having processing text in your mind.
>>539855878the grift begins by calling it AIthats where the entire bubble comes from
>>539855489most of the railroads laid are abandoned, same with the canals dug, the radio towers erected, etc etc etc. This is how it always happens, they bet on all the contenders, they win big on the winners, it works out. There aren't horrible consequences, there are just a few unhappy bagholders and some ghost towns who bet everything on being the next hub of whatever tech and everyone moves. The few companies that will have the pieces in place to train and license out desirable models will bring in big profits. >>539855598nvidia spark, take their pentaflop compute number with a grain of salt, pretty sure thats on 4 bytes not 64 bytes. Look on youtube, you can see examples of genAI video stuff as they scale up model size. No idea if it's linear, I think it would all depend on the model, the training data, the application, etc. >>539855643>they have a very high intelligence value, but are not consciouscompletely agree. As to the rest of it, we can't even define basic things like "story" let alone "conscious" so I'm not concerned with that all. Will it be dangerous? Maybe, sure why not. crossing the street is dangerous, I still do it because its useful. Dropping further into your asmov rabbithole. Silicon is very fragile, "they" won't go to war with their creators, as bio-based maintainers ensure their survival in case of a emp event. A hyper intellegence would recognize the importance of redundancy.
>>539856147>just keep giving AI money for 18 months and you'll totally seewhat impressive results are you talking about? that cartoon of will smith eating spaghetti?
>>539848891We have known this for years. Which is why people with inner monologue are actually more retarded than those without. Those without are processing much faster with pure intuitive logic than sounding out entire sentences in your head would allow
>>539856191>There aren't horrible consequencesCanals were beaten by railroads, which were beaten by cars, there was a progression of technology that fairly smoothly transitioned from one point to another here, and there were always people left behind.A much more accurate analogy is how everyone used to be on a small farm and we've gradually migrated to the city, for this there were horrible consequences that took decades of riots, striking, regulations, a couple wars.Some generic infrastructure project like a canal can't be compared to the largest publicly held companies in the world all developing the same technology which inherently most of them won't profit off at all, we're talking about pensions disappearing and the economy shitting itself here, not like a canal at all.Comparable to a shift from rural to urban because of the massive economic moves that would have to be made.
>>539856138I don't think you'd recognize your own asshole if it came up and shit in your face, anon. If you're just going to act like a fucking reddit pesud by *trying* to reframe the context to avoid arguing with the premise, i'm not going to engage you in good faith but instead just call you a retard, you fucking retard
>>539855619Conceptually, sure I'm sure something like this will exist soon, but if you read that article, that is 100% an enterprise system being designed for data centers, it is NOT a consumer product. Because even in the impossible chance something like this ever made it to the consumer market, the actual memory industry is fucked to hell right now, with DRAM prices sky high because all of the new production memory is pre-bought by LLM companies leaving none for consumer cards and sticks. So even if this were slower DRAM and not hyper-stacked GDDR, each of these cards would be like, $30,000.
>>539856349No, I understand these tropes and the theories behind them, all of which were dogshit sci-fi tropes about emerging AI.You might not, maybe you didn't read the books these were all originally pulled out of, but all of those were fiction.>Engage the premiseThe premise is from a sci-fi book and is meaningless words. It's hoping for some point where we make the robots do meaningless retard tasks enough until somehow an AI "emerges" that nobody can even make a basic simplified computer model of (not even with any actual points of consciousness at all to show)Basically show a simulation of this and then you'll have something, don't hit me with words.
>>539848891they're probably studying niggers who cant even think. AI is insanely intelligent I would say its working just fine.
>>539856191Yeah it's very similar to the forest fire problem - it has to do with the math behind chaos theory and potentiality but basically you end up with a highly intelligent, super dense forest you can walk around in and showcase - "look no fires, not even a history of fires 0 fires ever" to prospective investors, and they're not lying - there really has been no fires yet. But, mathematically speaking there is this hidden, massive potentiality for a catastrophic fire (well not catastrophic but more totally fantastical) and nobody has even able to correctly define yet what a spark or lightning strike even is, or how the physics of them works. It gets even deeper than that, but that's essentially the core of the argument and it's not just wishful thinking, it is *literally* the same math that governs the electricity that runs the fucking thing (funny how that works, no?)
>>539856349IT would be very disturbing and telling if someone recognized their own asshole
>>539856439go eat some corn or something you fucking cattle, im done here
>>539856564lel i kekd thank you
>>539848891>has nothing to do with AI>NOOO IT IS ABOUT AIdeath to all afro-columbian monkey men.everyone already knew this, that's why binary and prime number correlation exist in all languages independent of using literal words.
>>539848891NPC cope lmfao
>>539856689Its npc cope about not having a soul>we are still human regardless of our soullessness Basically
If AI can't create a solution or invent a cure/innovation/hardware in 5 years then pull the plugI'd prefer skynet to deepfake porn and filter apps
>>539856338>there were always people left behind.The canal monopoly was outperformed by the railroad monopoly which was outperformed by the car monopoly (except for heavy freight). Nobody talks about all the speculative infrastructure collapses
>>539848891While the mechanisms for logic may be separate from language, language is what makes logic useful. Even an animal can use logical reasoning, like a crow figuring out how to get a crumb out of a jar. But language is how logic is put to use in the world, for language is the only way to communicate logic to others in an efficient manner.
>>539851937>Don't blame me blame capitalism for being a shit showAre you saying that Karl Marx would have been able to invent robot cat girls for us by now? If so sign me the fuck up I'm going to my local DSA office tomorrow morning.
>>539848891Just because it learns by consuming language and has an output process of predicting "the next word", doesn't mean it doesn't internally represent logical reasoning.
>>539851069AI is already superior to 99% of jeets and Apefricans working for the government.
>>539856338it was not smooth, canals were being built long after the railroads were booming. Cars, trolleys and horses co-existed on the roads for decades. Very few companies are "all-in" on AI. Many companies are dabbling in throwing some cash at using it, not developing it. Just some tools to help their middle managers seem competent. They spend more on team building retreats. Some companies spent a ton of money on video phones and other telephonics pre-internet and they survived just fine. This is very normal. Pensions won't disappear because openAI and oracle go bust. Most people don't have pensions anyway so its a moot point. >>539856393>it is NOT a consumer productits a 240W pcie card, You could throw that in most towers. And you'd be able to buy used one at auction in a few years on the cheap. Or just rent some time on one.DRAM going skyhigh, more build capacity being put in place, and the AI bubble popping would mean a ton of cheap RAM on the market. I don't see the problem. Outside of niche applications like llms and 8k video editing, its not like you need a better computer than what $2k got you in 2016. You'd still be able to play every game that gets released. There are basically no programs that take advantage of multi-cores, chip speed has been largely stagnant, and the instruction sets have barely evolved. A decade old $100 laptop is fine for 99% of the population. Why are you stressing about RAM prices? They don't matter at all. Worry more about hdd's getting jacked up, since that actually limits what people can share, store and host themselves. Renting compute is cheap and easy.
>>539849295Retard. Fucking midwit reddit retard.
>>539848891Indians everywhere are sweating bullets right now
>>539856575Yeah zero argument every time>No simulation not even a simplified one>Just repeat sci-fi seriesJust one, just one singular simulation that shows even a mote of said consciousness, I'm waiting.>>539856876>Most people don't have pensionsIt's the opposite, an ever-growing number of people's pensions rely on these publicly traded companies. Social security is a pension that also does rely on said companies.
>sorry Chuddie it seems the brain is not necessary for cognition or consciousness, back to the drawing boardThats how retarded OP is
>>539848891LLMs don't use words either, in the simplest form they use tokens, which are not bound to a single text representation and can be shared across multiple languages.but in reality they use token clusters which point to areas in very high dimensional space.
>>539848891I'm banned on x but can someone reply to this low IQ foid bitch with:That's a pretty low IQ, non-sequitur conclusion. Assuming for the sake of argument that LLMs completely diverge from human brains' mode of thinking, that says nothing about what they can and can't do. Wheeled vehicles are not bipedal and yet they're better than you at transportation.
>>539856807Well China did and it's a communist country
>>539856191>we can't even define basic things like "story"Let me guess you think there is no definition for a "woman".
>>539856876>its a 240W pcie card, You could throw that in most towers.Yeah, no shit. I'm saying it's not for sale and you can't afford it.> and the AI bubble popping would mean a ton of cheap RAM on the market. We'll see. In my experience having lived in the real world, prices never go down, ever. This is just the new baseline. >Why are you stressing about RAM prices?Because I work in Photoshop and Blender while also running Chrome, which means that my 64GB of RAM is barely enough to run that and not kill itself. >Worry more about hdd's getting jacked up, since that actually limits what people can share, store and host themselves.Thankfully, I invested in a few 12TBs a few years ago, so I'm pretty set on that front.
>>539856322True
>>539856990oh, I get it. You get a quarter or something every time you get a (you)>You're retardedThat's out of pity, anon. I pity you.
>>539854757good summary>>539850088>AI>its the biggest lie in economic historychecked, Hans gets it
>>539853372really?
>>539856990social security has $0 in the stock market. They are prohibited by law from this, pretty sure they are allowed to buy some government bonds. I'd be surprised if 20% of the US had pensions. Yeah most people have retirement funds (those aren't pensions) but a fire sale on suburban mc mansions isn't exactly stressing me out.
>>539853477User 11:13PMGenerate some controverisal AI-related images for this post. >>539848891Model 11:13PMI can't generate images of controversial AI content. Would you like me to try generating a different kind of image instead?4.4s
>>539857150>It's like a fire in a forest bro, we don't know what a fire is or even what lightning is but we definitely know that we're making the forest right and something will certainly happen over timeThis isn't an argument. These are words.>>539856563see that, it's not an argument.>>539857223>0 dollars in the stock marketMoney going in / out of social security depends on the economy providing people with jobs - public companies do this. Stock market crash = suddenly all the money from social security tax isn't coming in.In the past the average person wasn't invested in the market to be hurt by any of these crashes, but if they had retirements of any kind or a job they were at risk of losing it and not finding another for years - this is very bad for any existing pensions.
>>539853722>when Islamic terrorism still existsYes! See "The Geopolitical and Legal History of White Warfare Against Enslavement" https://write.as/mwq381bw4r7ms.mdARTICLES OF IMPEACHMENT against Chief Justice Roberts https://write.as/q1jc2dzd56xhp.mdDraft Petition for ReHearing in Trump v. Barbara https://write.as/edit/6j8azq11mfs94
>>539857095When I worked in premiere yeah 64GB felt like it was just barely enough lmao. Okay I feel your pain. Adobe programs blow and yeah thats a 1% "power user" catagory. You can offload that compute though. On premiere adobe had a 1st party tool to link in a second computer, I had a $400 workstation in the closet that did my renders. Blender has tools to do the same, allows you to keep working rather than getting jammed up looking at a loading bar. That second device can be total shit, you'll still save time.
>>539854126>there is no AI>>539854182>for free>>539854363>maintaining a critical dynamical regimeSee "The Geopolitical and Legal History of White Warfare Against Enslavement" https://write.as/mwq381bw4r7ms.mdARTICLES OF IMPEACHMENT against Chief Justice Roberts https://write.as/q1jc2dzd56xhp.mdDraft Petition for ReHearing in Trump v. Barbara https://write.as/edit/6j8azq11mfs94
>>539851341This.>>539851901This.The right hemisphere doesn't talk back but but it computes beyond words, without it you wouldn't understand metaphor, poetry etc.
>>539854757>They don't "think", they don't "reason", they never have.This is not correct. As the layer count of LLMs increases, they can resolve greater amounts of different concepts and language options. Thus, a coding LLM can know before it write the first output essentially what the last "main" section will need to call. See "The Geopolitical and Legal History of White Warfare Against Enslavement" https://write.as/mwq381bw4r7ms.mdARTICLES OF IMPEACHMENT against Chief Justice Roberts https://write.as/q1jc2dzd56xhp.mdDraft Petition for ReHearing in Trump v. Barbara https://write.as/edit/6j8azq11mfs94
>>539848891There are countless signs and symbols that are universal and understood by everyone. Languages do make it easier to communicate compares to symbols alone.
>>539857195Alphabet reported its first cash-flow-negative quarter since its 2004 IPO, with Q2 2026 free cash flow at -$5.9 billion after capital expenditures doubled year-over-year to a record $44.9 billion. CFO Anat Ashkenazi raised full-year capex guidance to $195–$205 billion and confirmed spending will rise further in 2027. About 60% of infrastructure investment went into servers, primarily TPU-based systems. Google began selling TPU hardware directly to customers' data centers for the first time, with inventory quadrupling to $10 billion. Alphabet issued $49.6 billion in stock and $20.3 billion in bonds, pushing long-term debt to $98.2 billion. Google Cloud grew 82% to $24.8 billion with a $514 billion backlog. Key customers include Anthropic (up to 1 million TPUs) and Meta in talks for multi-billion-dollar deployments. The seventh-generation Ironwood TPU scales to 9,216-chip pods rated at 42.5 FP8 exaflops.https://daily.dev/highlightsSee "The Geopolitical and Legal History of White Warfare Against Enslavement" https://write.as/mwq381bw4r7ms.mdARTICLES OF IMPEACHMENT against Chief Justice Roberts https://write.as/q1jc2dzd56xhp.mdDraft Petition for ReHearing in Trump v. Barbara https://write.as/edit/6j8azq11mfs94
