Has AI plateau'd?Give it to me straight.
>>106530631Yes, hardware isn't improving in a rate where AI development can keep going at a paste it always has.
>>106530631Wait for Grok 5. Its gonna be released end of this year.
>>106530657And... it's going to be unaffected by diminishing returns, why exactly?
>>106530631It’s still improving, but the problem is that it doesn’t generate money. Everyone except for NVIDIA is operating at a loss.
No, but big corpo-ran AI looks like it's slowing down drastically. We'll need new architectures, or even to abandon the LLM framework for something better.
>>106530674>No, but yes
>>106530664Thats what they said about GPT4->5. But Grok 4 proved them wrong. Huge leaps were made. Grok 5 will be the true test of scaling.
>>106530685Still not explaining why
>>106530756Because the scaling hasnt stopped for xAI. It has only stopped for ChatGPT.
>>106530773>Because the scaling hasnt stopped for xAI.Elon propaganda.
>>106530792Consider yourself why you castrated yourself first
>>106530799rent free
>>106530799>>Consider yourself why you castrated yourself first>you have a different opinion then me this must mean you ether are a jew or trans.I have no idea what "castrated" is supposed to translate to but I am so tired of people like you.
We don't know. It seems we are reaching diminishing returns, but a new paradigm may be around the corner, given the alternative is the whole industry crashing down and maybe the Chinese dunking on the West given how hellbent they are on getting this done. AI is strategic now, so I guess there's plenty of room for surprise.
>>106530908not that retard but you don't know what castrated means?
>>106530934He's just tired, but he meant that something may be lost in translation due to the presence of undesirables in this God forsaken Mongolian basket case of a forum.
>>106530963in... translation? wut? We're all speaking English here
>>106530908>i castrated myself over difference in opinionkekYou will NEVER be a woman. You troons lost.
>>106530657Elon said 2 more weeks
>>106530631Graphene processorsNuclear batteries
>>106531614And in how many weeks? 2?
Interesting how all these "huge leaps" are only seen in marketing products and not in actual usage.
still literally never seen one of these monstrous "ai" models do anything useful
>>106531660They're great for learning things. I guess learning is just not something you do
>>106530668Why are they profitable from it?
>>106530631>Give it to me straight.nah sit on all fours and take a deep breath to let daddy in
Imagine the things the authorities will do with AI to frame those who speak out. This is why they want you to scan your face for accessing mature web sites. They could put you in a video of raping someone. It's quite scary. Anyone else worried about the future of AI?
>>106531704good morning saar, I too wish to know the needful to obtain dollar from AI saar
>>106532244>This is why they want you to scan your face for accessing mature web sites.Only if you are English or American. Remember the rest of the world doesn't fuck with that shit.
>>106530799good morning saar
>>106532311Other countries will all follow eventually. The UK is the test phase. You go outside? Your face is being captured thousands of times by cameras. Zoomers post thousands of photos of themselves every year to public social media. Fighting for privacy will mean you have something to hide when the masses accept this new way of life.
>>106532359The UK has a modern surveillance apparatus for more than 20 years now. Same for the US, but with more sophistication due to the big tech arm. These are outliers, and very hard to copy without the same level of investment in the tech sector.
>>106532359They won't even be this obvious about it in the future. It will be built into every CCTV across the country. There is barely anywhere in most UK towns and cities now without a camera capturing every street and open space.
>>106530631>Has AI plateau'd?Obviously. Even if you didn't see where it's all going watching them increase the parameter count and compute requirements by orders of magnitude for marginal gains in fake benchmarks, it's clear that they have absolutely no idea how to make their token predictors actually reason, complete tasks reliably, stop hallucinating, provide accurate information, admit it when they don't know something etc. They were waiting for these properties to magically emerge, but it's not happening no matter how much bloat they add.At the same time, they're completely invested in their failed assumptions and false promises to backpedal and rethink things fundamentally.
>>106532244Authorities dont need AI when they have media, journos, institutions pushing the WEF's tranny death cult propaganda. AND attacking anyone that disagrees with it as nazis. The mass of the population gets culled into propaganda and think they're real smart/rebellious by castrating themselves and advocating for censorship
>>106530631what the fuck is the source for that webm anyway?
>>106532471>picthat's not how AI works at all tho
>>106532491>that's not how AI works at all thoIt exactly depicts the core of "deep learning". Your fancy model is still a variation on that, with a computational flow that conforms to the pic.
>>106532491That's how training goes, in a rudimentary way
>>106530631No, industry leaders still predict we'll get AGI in 2027
>>106530987No we're not, some are speaking Indian english like the pajeet who said something about 'castrated' whatever the fuck the poo even means
>>106532552Wrong, and you don't know shit about programming
>>106532599...you seriously dont know that word?
>>106530631fuck no, that guys face looks retardedreminds me of said image
>>106532623the webm in the OP is not AI
>>106530664It's made by papi.
>>106532545Good. Can't wait for the bubble to burst.
>>106532648you are AI created by the government to cover up AI
>>106532478>skullfucked /pol/troon knows nothing about the subject, so it starts barking its favorite buzzwords and foaming at the mouthWhy even come here? You can do that back at your tard ballpit.
>>106530631Damn anon that's creepy as fuck XD
>>106532610>WrongAnd yet you are unable to point out anything specific that is "wrong" about it. You will fail to do so in your next niggermonkey post, for instance.>you don't know shit about programmingI know enough about programming to convert any neural net into exactly that structure. Seethe about it. :^)
>>106532679Malding tranny. You commies lost.
>>106530934>>106530987>>106532616SAAAARRRRRR GOOD MORNING SAAAAAAARRRR DO THE NEEDFUL
>>106532745Stop crying, just leave.
>>106531704Because NVIDIA doesn't sell AI, they sell shovels.
>>106532762Seethe more troon. Your commie ideology lost. The west is being reshaped to fix you troons
>>106530631Kind of yeah. It's gotten better at orchestrating tasks and holding a larger context window, but it's still retarded and still hallucinates the same shit over and over. Ask it to help install shit from the cli and it recommends package versions that don't exist despite having access to the Internet to go out and verify the versions and dependencies.
>>106530918>a new paradigm may be around the cornerCope
>>1065306854 to 5 was a decrease in almost every way that matters.
>>106530631In some measures it has plateau'd but that also doesnt mean its gonna stop advancingIts already at a place where it is a real threat to entry level jobs
>>106532782>no call me baste i want muh /pol/ echo chamberJust go, this isn't working out.
Transformer AI has.Until someone makes a better model, it will forever be fucking around with values that maybe give slightly better generation times, maybe beat a pointless benchmark and cost a mansion to do so.Deepseek was the only major change in recent times that resulted in a huge boost to generation speeds.Chain of Thought isn't revolutionary - half the fucking time the models ignore their steps, and I was doing that shit manually way back on Character AI, it's the same damn model going schizo with itself but automated.I can typically get better outputs WITHOUT the CoT being on, which is why ChatGPT5 sucks particularly bad due to its automatic mode switching. They really fucked up with it.Don't even get me started on AI Agents, most of the time they ignore their own fucking API calls or send back malformed shit, then have catastrophic breakdowns.
>>106532795>it's still retarded and still hallucinates the same shit over and overIt literally doesn't matter.If there is even a 0.01% chance that it could be wrong then it it useless because that means that all output needs to be validated for anything that matters.The core issue is that a machine can never be held accountable when it tells the button pusher monkeys running it to eat 500 grams of Tylenol instead of 500 milligrams and someone ends up dead from liver failure.Because of that it doesn't make a shit what AI can do, if it can't produce verifiably correct results 100% of the time then it will only ever be used for shit that doesn't actually matter. And news flash, shit that matters are typically the things people are actually willing to pay for. Your blogspam shit website doesn't matter, your stupid newsletter and spammy marketing materials don't matter, doctors matter, lawyers matter, engineers matter.
>>106532880Exactly. At least when the human tells you something patently wrong and retarded you can almost guarantee they won't come back 5 minutes later to suggest the exact same thing. And if they did you would just fire them instead of dedicating another power plant worth of energy to see if they become God.
>>1065328095 is better for me. But I suspect it's a result of them benchmaxing. So it's better for how I use it but less intelligent overall.
>>106532920The difference is, when the human tells you to use 2 inch nails instead of 6 inch nails and your house collapses, you can sue that human for damages and fix your house if you win.You have recourse.This is exactly why that judge told those lawyers who tried to use ChatGPT in the courtroom to fuck off and never come back.Assurance has value, accountability has value, AI offers neither, it is completely unacceptable for anything of real consequence.Only people cashing in on stupid VC morons would say otherwise.
>>106530631The main use case for AI is to sift through the bulk data collected by corporations and fed to government agencies to help profile every living man on the planet. For that purpose, it is probably more than good enough by now. Otherwise there would not be such unprecented investment in datacenters.
>>106532967>The main use case for AI is to sift through the bulk data collected by corporations and fed to government agencies to help profile every living man on the planet. For that purpose, it is probably more than good enough by now. Otherwise there would not be such unprecented investment in datacenters.This is the objectively correct bottom line that should be the start and end of every "AI" thread. They don't need thinking machines. They need better statistical models of their human cattle and social dynamics so that they could manipulate and control everything better.
>>106532359>Bitch about chinese surveilance>Ends up copying chinese surveilanceWhat did westerners mean by this?
As a field and no. As a trend, maybe.
>>106532997Rent free
>>106532727>spouts complete Dunning-Kruger nonsense>"nuh uh, spoonfeed me!"Fuck off retard, do your own research
>>106532754Why is /g/ consistently the lowest IQ board on this site?
>>106532993Why did you quote the entire post? Fucking idiot
>>106532244Just use an AI generated face.
>>106530631AI is better than ever and it is improving faster and faster. Don't listen to the haters. In six months the world will be unrecognizable to today. AI is an infinitely growing beacon of prosperity.
>>106530799Holy ESL grammar.>>106530987Are we? Are we really?We need one of those negro voting tests from Jim Crow era Southern states to filter out the brownoids.>>106530918>a new paradigm may be around the cornerThis faith in absolutely nothing but dreams is quickly approaching cult/pseudo religious levels.>>106531660Useful applications of AI isn't in LLMs. It's in highly specialized applications like genetic sequencing. The ability to better understand pharmacokinetics and pharmacodynamics on a per person basis is going to be important. Medications can be tailored to an individuals genetic make up; applications of previous medications for new diseases due to better understanding of how they interact with the body, etc.
>>106533136Really? Because the new models haven't improved as much as the previous models did. I smell diminishing returns
>>106533136>Improving faster and fasterOverall maybe. Like improvements in video or voice or 3D renders or whatever keep coming out that could really disrupt creative spaces. But in areas that actually matter where real work gets done you're still left with vibe coded security defects. Basically for every creative application of AI there's always some level of imperfection that you can kind of just ignore. But those imperfections in software are a banking error or incorrect medication or a reactor exploding. Basically it will disrupt the artcels and OF whores and force them to get real jobs but unfortunately you're gonna have to keep waging for a while.
>>106533131>Why did you quote the entire post?Because of this:>>106532993>This is the objectively correct bottom line that should be the start and end of every "AI" thread. They don't need thinking machines. They need better statistical models of their human cattle and social dynamics so that they could manipulate and control everything better.
>>106533114Notice how I correctly predicted your inability to point out anything "wrong" in that pic. Do you need me to spoonfeed you how to code a function that converts a neural net into a tree of conditionals like that, or do you want to remain the saddest victim of Dunning-Kruger ITT?
>>106534109>cars work through magic>"no they dont">HAH YOU CANT EXPLAIN HOW CARS WORKNot doing your own research for you.Do it yourself, or keep flaunting your retardation in public, I dont care
>matrix multiplications have exponential complexityWho's gonna tell them?
>>106534221Guess I really do have to spoonfeed you. Ok, here you go:https://pastebin.com/raw/3At3KBcWI wonder what kind of delusional cope you're gonna cope up with against a direct, runnable demonstration of the equivalence between neural nets and an if-else tree.
>>106534342Whom are you quoting? Are you having another one of those episodes?
>>106530631At some point there will be a combat between a Tier One and Two country and this will be when everyone literally shits their pants and starts calling for an end to AI this WILL happen.The scenario is USA versus Russia lets say. I dunno maybe Russia is too strong and would use Chinese AI. So say USA versus... a random country that has jet aircraft.AI will be able to predict every plane movement, all missiles, hit chaff at perfect timings, fire missiles with perfect timings. USA will shoot down 180 Indian jets and India will shoot down 0. Then the time of AI has finally come.
>>106534406You have chosen to keep embarrassing yourself
>>106534439We can already do all that shit and don't need AI
>>106530631damn the new doom looks crazy
>>106530799>Consider yourself why you castrated yourself first...what?
>>106530631Honestly, I think AI can advance but... it feels really solved to me. How much better can it get?The women on Grok look really realistic to me. I have seen on /gif/ its basically 10% less real than reality I don't see much difference its just a big fat tittied lady getting dick. AI videos are just like oh wow heres some humans walking around. la de da, no different than say computer graphics in a movie. Call me an idiot I dunno it really seems as good as it should get. Yes even if you wanted too you can get any kind of girl do a 360 and see her pussy, asshole, holes in her nipples, skin bumps. Its all possible now.Maybe its just the polishing phase and there is no need for "revolutions" its just actually extending the time of video footage and making less mistakes.I think this means plateaued.
>>106534602SELL SELLSELL SELL SELLSELL SELLTHE NORMIES CAUGHT ONTO THE SCAM
>>106534537>the biobot's chosen coping method: a fully generic non-responseGuess I really do have to spoonfeed you. Ok, here you go:https://pastebin.com/raw/3At3KBcWI wonder what kind of delusional cope you're gonna cope up with against a direct, runnable demonstration of the equivalence between neural nets and an if-else tree.
>>106534786>My 10 layer toy model is the same as a 70B parameter model>I don't know what scaling isYou could dig a hole to the center of the earth but hand but it might be quicker to use a machine. Not the anon you're responding to btw. Just thought you're either retarded or autistic. Either way.
>>106535188>10 layer toy modelThat transformation has no cap on the number of layers. Try again.
>>106534602I'm in those video threads as a creator rather than a beggar. It's good enough to goon for sure but it's still not as good as it needs to be. Following instructions is still luck of the draw unless you're willing to attach 30 Lora's and controlnet the entire thing.
>>106535454>No capOk. Run it on a 70B param model and show me the output. I'll see you in a year when your response is ready.
>>106535576>Run it on a 70B param model and show me the output.Run a 70B param model on your computer and show me the output. What is your argument? You seem to be literally retarded and incapable of abstract thought.
>>106535630My point is that scale is an issue. Just because something is possible using a dumb method doesn't mean that's the way it should be done. Be honest, does your neighborhood look like this, saar?
>>106535712>My point is that scale is an issueHow does this issue apply here? Pretty funny how much you struggle to go beyond vague rhetorical patterns, just like your new "AI" deities.
>>106535712>>106535753>no responseYep, that's a concession. The scaling issue doesn't apply to one representation any more than the other. They scale the same. It's the same computational structure.
>>106535753It applies here because your original argument is "LLMs are just a big decision tree" and "I can run a kiddy script that turns a 10 param LLM into a decision tree". While you're technically correct, in a practical sense it's not viable. Can you technically pile up a bunch of chairs on top of one another and reach the moon? Yes. But the energy required far exceeds the effort to just build a rocket instead.
>>106536073Easy on the addy, kid.
>>106536097>in a practical sense it's not viable.Why did you backpedal? You said it was about "scale". Both implementations scale exactly the same. Also, why is it "not viable"? Make sure your explanation is relevant to the substance of the asserted equivalence. Protip: "it's not laid out optimally for my specific hardware" is not relevant.
>>106536273Are you ESL? It's not viable BECAUSE it does not scale. Obviously I don't have the hardware to run gpt on my machine locally but I can run a smaller 8B param model locally. Running your script on a similar model to output a decision tree would still take ages. It's not a viable replacement because when you're using software you care about output AND time to receive that output. You've got to be Indian. Nobody can be this retarded.
>>106536378>it does not scale.Do you actually understand what "scale" means? Protip: it means something more specific than "if I make the model bigger it will be too much for my hardware". In your next post, demonstrate that you know what it means. Protip #2: you will refuse to do this and back out with some excuse about not "spoonfeeding" someone who is clearly 3-4 stdevs about you in IQ.>It's not a viable replacemenWhy did you backpedal? You said it was about "scale". Both implementations scale exactly the same. Also, why is it "not viable"? Make sure your explanation is relevant to the substance of the asserted equivalence. Protip: "it's not laid out optimally for my specific hardware" is not relevant.
>>106536460I've decided I'm not going to waste my time explaining things to a third world retard. This conversation doesn't scale, sweatie. You can call this a victory or whatever.
>>106536526>I've decided I'm not going to waste my time explaining things to a third world retard.Notice how predictable you are. Called it right here:>>106536460> Protip #2: you will refuse to do this and back out with some excuse about not "spoonfeeding" someone who is clearly 3-4 stdevs about you in IQ.Maybe some other retarded niggermonkey can pick up from where you left off and explain how it's not viable due to "scale". Except they can't. It's precisely viable due to scale.
>>106536591due to scaling*
>>106532809I never paid to access the o models but 5 high is better than all the free 4's.
>>106530631How long do you want to bet on? A year? Ten years? A hundred years?
>>106530685>Huge leaps were made.they weren't.
>>106536591You show that benchod!
>>106536723I literally did show that. You're just a retarded American "white" niggermonkey, dumber than any shitskin, so you're oblivious to this. Why do "working class" retards from /pol/ even get so mad about curryniggers taking over the American IT sector? You don't cut it for any kind of IT job, even with the abysmally low standards prevalent in that globally despised, third world cleptocracy you call a country.
>>106536808>I literally did show thatShow what, ESL monkey?
>>106536841That it runs on current hardware and it scales, niggermonkey. Try to keep up.
>that one self-hating jeet who's seething because he brought up scaling but then got stumped by someone asking him to explain what it isthe final state of this board
>>106532471/thread
>>106530631Nah, it's getting incrementally better. The bigger problem is Moore's law is plateauing and has been for some time, so hardware is going to be a problem for some time. If this was the 90's or earlier you could just wait three years and a £$2000 desktop would manage easily what previously you needed a server cluster for.
>>106536859>You show that benchod >I did show that>Show what>That blah blah blahI'll give you a hint since you're an ESL retard. "You show that {so and so}" is not a call to action. It's a cheer to say "wow you had a really strong showing against that {so and so}.Hit the books and maybe the shower.
>>106530631That's a real video of Kier Starmer standing in a tv studio btw
>it's responding to itself now
>>106536962>80 IQ American semibrownoid doesn't understand how language can be ambiguousNo point talking to nonwhite races. Closing this dogshit thread. But you WILL reply again, just like your subhuman programming compels you to.
>>106537018>There's ambiguity in the common phrase temple "you show {subject}"Maybe stick to arguing the semantics of scale. English grammar isn't your lane.
Just give it two more weeks and another trillion dollars and it will break containment, trust the plan
>>106530799stfu you poojeet faggot
>>106530631>Feed AI "knowledge">Can't actually feed it all so it has to be a glorified google researcher for what doesn't fit.>include biases as axioms (don't be racist, don't blame the jews, we're all equal) that ruin the basic capacity of understanding cause > consequence>police content with those retarded precepts in mind so there's now actual gaps in knowledge and wild logic leaps in the training data.>act surprised when no matter how much training and knowledge you give to the AI, it's still a lobotomized, inhuman 'tool'.>"wOW GuIsE HAs AI pLATeau?!?!?!"The closest to a proper conversational-level AI was mechahitler.
>>106537262Honest question: what do you use AI for?
>>106537262>there's now actual gaps in knowledge and wild logic leaps in the training data.That would still be the case if you stripped out all the wokeshit and would be doubly the case if it all got replaced with your retarded ideology.
>>106530631Dudes turning like cardboard
>>106532483Some news station trying to be cool while covering the UK elections.
>>106536974Can't fucking find it for the life of me. I guess all the psyops around this guy are suppressing that vid
>>106537330Doom 2025 went back to its roots and uses billboard sprites for the demons. The only thing they fear is (You).
>>106537366Wdym is it like an LoD thing?
>>106537262>Get a message from a recruiter on LinkedIn >Indian>Look at their credentials>Some Indian university I've never heard of>Ask gpt if it's a legit school or a degree mill>No it's a legit public school>Ask if they've ever had any cases of hanging out degrees without proper academic integrity>Lists out several cases of fake professors, pay for degree scandal, or professors changing test answers to inflate grades>Point out the hypocrisy >You're exactly right but here's why that's not a degree mill >Ask if it can name a time where my own university had done anything similar>Brings up some bullshit about students taking a knee for the national anthem >Completely unwilling to admit that India's entire academic system is compromised if a school that's been caught multiple times handing out degrees is still "accredited"The people training these models have their own politics and they've hard coded them into the models. Ask it a problematic question with an obvious answer and you'll see it every time.
>>106537432Well duh, why the fuck would you use AI for political shit you already know? Is that what you'd use AI for? Just to validate your beliefs?
>>106530773> lame unsubstantiated propagandaHi Elon
>>106532471>watching them increase the parameter count and compute requirements by orders of magnitude for marginal gains in fake benchmarksBINGOthis is the key
>>106533136>. In six months the world will be unrecognizable to today. AI is an infinitely growing beacon of prosperity.yo Sam Altman, stop visiting this board and keep sniffing more flue
>>106530908>I am so tired of people like you/g/ is full of reactionary ESL zoomer jeets that seethe over everything
>>106534553>We can already do all that shit and don't need AIcorrect
>>106530631data and RL environment wallworst timeline increasingly likely: no singularity, total wagie death
>>106537286Discussing politics so I can better know how to argue against libtards.
>>106537688They defeat themselves, there's no need to argue
>>106530908>>106532337>>106533159>>106534597>>106537176lmao newfags. lurk moar.
>>106537494You dumb jeet. The core of my initial question was "can I trust this random jeet in my inbox" and AI responded "yeah they're totally trustworthy". If I didn't have the common sense to dig deeper I'd be neck deep in shit right now. These biases aren't as simple as little political questions. They're philosophical assumptions baked onto the model that if blindly followed will turn my country into Mumbai. But you know this.
>>106538204>can I trust this random jeet in my inboxAgain, why the fuck do you need an AI to answer this question? Are you retarded?
>>106538239If it's this obviously wrong at first principles what hope does it have for higher order thinking?
>>106532359>>106532396damn, i really wanted to move to the UK. but like, late 90s - early 2000s UK with sane internet and goths around. does it suck that much now?
>>106530657>>106530685this sounds like what twitter's shilling department (or perhaps elon himself) would sayhave you considered buying an ad?
>>106530631We just need another 500 trillion and the entire electrical output of the US and EU combined and we'll have agi in two more weeks i SWEAR
