Your company can run the most powerful AI models locally without having to share any data with Open AI or Anthropic.https://youtu.be/vDmjw-c_R2k
>>545705563did you know nvidia will pay you $1000/month to run a local ai at your house? Why would you build one yourself when its free?
This is what the kikes like scam altman do not want. This is why they conspired to increase PC part prices, to put this out of reach of the average person.It's been proven time and again when the people of ambition out in the general public are left unfettered to figure this shit out they do better than bajillion dollar corporations.
>>545705563Buy and ad Linus.
>>545705628>take money and hardware>claim connectivity problems>use it locally>if they want their hardware back you claim you already sent itthey have no actual humans to collect their stuff again and after months they will give up.
>>545705716I'm not Linus, I'm just a guy who recognizes the cultural and political significance of his 30 minute video teaching people how to implode AI companies' /profitability.
>>545705563Did you guys know that you can just take a big fat stinky shit directly on the floor? Then you don't need someone else to do it for you!
>>545705563Are local models good enough to recompile vidya like Claude?I have no other needs for this atm
>>545705797nigger detected
The fact that my pos pc can make.videos on par with grok is absurd. Like yeah grok is faster but the quality is shit and censored. Really hope the ai bubble crashes hard and dems.block ALL datacenters so hardware will come down to reasonable levels again.
>>545705628You pay nvidia $150 a month and they pay your electricity bill or something like that. The average homeowner would net like $1k in savings a year, not per month. And just like every big tech pilot program, it's only available in a handful of states near california
>>545705563> 30 minute videoAt least post a summary, you lazy nigger.What HW, how much does it cost, how fast it is on which concrete models?Any specific SW? Etc.
>>545705825Thats EXACTLY what someone like Linus would say
>>545705865no
>>545705851Sorry, but it's not comparable. No one saves your shit and uses to improve their sewage system. No one depends on scooping your shit to claim profitability and there are no sewage companies receiving trillions of dollars government and private investor investment
>>545705947Why don't you use AI to transcribe and summarize the video for you?
>>545705825>AI companies' /profitability.They're not profitable, they're getting infinite funding from taxpayers because it's a giant jewish scam. No logical argument will unseat their hegemony, because it's not based on logic to begin with
>>545705563ai is fake an gaythe cloud is web hostingwhite people believe all marketing
>>545705947>At least post a summary, you lazy nigger.https://www.lttlabs.com/articles/2026/11/02/jarvis-at-home
>>545705865open source models can help with coding and decompile software, but not as quite as good and not nearly as fast as the corporate stuff. It can get the job done though.
>>545705563how does that work without being hooked up to big data?
>>545705825Ok Linus
>>545705563The only thing bursting AI bubbles are the complete blue wave victory a few days ago, Nvidia are in freefall, data center shutdown talks are being had, and Elon wants to move
>>545705563I'm only running local AI and offline information resources in case the internet goes out. I don't have any use for it otherwise. I need to download some books from project Gutenberg and survival manuals and medical stuff. Basically turning my laptop into a prepper device. Battery won't last long in the event of power outage but my laptop will be a wealth of information while it has a charge.
>>545706203Just get some solar powerd charging devices.
Can it suck your dick?
>>545706132LLMs are trained on a lot of data and you can integrate them with offline resources
>>545705689FWIW, during the ethereal crypto mining boom, amazon would bill about 7-8x what I made running GPU servers. It's probably worse these days. They operate at around 85-90% utilization (so only 85-90% of the time they're making money off their machines) whereas I was at 100% utilization (every second my crypto rig is running it's making money), but that doesn't come close to explaining the discrepancy. They also get cheaper electric rates, benefit from economics of scale with cooling and such, and get cheaper rates on cutting edge equipment (up-front cost is insane, but when you have computers running 24/7, even a slight efficiency improvement leads to massive cost savings). This is how much price-gouging there is going on, and if they aren't wildly profitable it's simply because they have millions of jeets acting as boat anchors.
>>545706052>They're not profitablePeople buying their shares seem to believe that they will become immensely profitable one day. But of course that can only happen if customers pay these companies to use their models. Here in Europe there's a lot of reluctance to share any data with those companies because they can steal your information. Also datacenters cause too much hazard. So organizations will increasingly start running their own local AI servers, making AI companies look much less profitable than they appear now.
>>545706132Other parties have stolen big data's models and distilled them into open weight models (or made their own) which anyone can download and run locally with a capable enough computer and a small amount of technical know how.
>>545706278no. srsly...what can you do with it?
>>545706312*ethereum, fucking autocorrect
>>545705563Is local AI coding good? That’s my only skill gap.
That jakkuh ex-sidekick of his … split off his own channel. A bit annoying but cute. Wish he was a gay bear cub.
>>545706069Thank you.I expected more than this setup. I'm thinking about buying a RTX PRO 4050 or some GB10 implementation.Unfortunately, small or overquanitified local models are unusable for coding. His setup could drive something more powerful, like Qwen3.8-next which is borderline useful for coding and inference.However, having a RAG at hand is also important for coding, and 24GB aren't enough for running both.> nb4 https://github.com/JustVugg/colibriNot there yet. Maybe a 2x 24GB GPU on PCI x16 is the most future savvy option, but then the cost increases even further with RDIMM etc.I'm afraid that it isn't possible to get usable local AI under $10K nowadays.
>>545706460>Is local AI coding good?Same quality as coding by Claude, GPT, or DeepSeek...
>>545706132Models are quantized meaning big models can lower their resolution to run on a shittier machine. You can't do it too much though or you'll get slop.Also sometimes is betting just running a different model non quantized that's focused on your specific task.Some models are mixture of experts so it's like a half dozen models working together and checking each other. The MoE is the best bang for your RAM out the box.
>>545706576>I'm afraid that it isn't possible to get usable local AI under $10K nowadays.It's not something that the individual consumer can afford, but for organizations it's very doable.
>>545706671>>545706576I got a 128GB unified memory for 4k It's too shitty to run intense models it's good to multi task mid models
>>545706714Do you think the shit this guy is working on is legit, or is it retarded?He seems to have massive speedups on shit hardware, but I can't tell if it's bs.https://www.youtube.com/watch?v=RLXjscqnp6A
>>545706365>open weight models>>545706311seems like it would only work for fairly specific jobs, but IDK or understand.What I'm bullish on is AI for Home Inspection and filing out new HUD UAD 3.6 Appraisals off a DIY owner's photo scan, and replacing the Realtor's $100billion yearly total commissions.Replace Realtor's with a bank-centric process where the first step is to have a bank pre-approve a mortgage for the house, not just the buyer.
>>545706754idk I just got an AMD and I'm a noob
>>545706330Google is banning entire school systems from using google search if they don't also adopt google AI assistant. Communism only worked with a gun to the user's head, too.
>>545706783i'm pretty sure antisemitism is illegal in your country, too
>>545706671But I am an individual consumer. I don't care about corporations, at work we have frontier subscriptions we can use for coding.I want AI at home. My AI. The idea of not having my own tools disgusts me.>>545706714What exactly do you have and how do you use it?
>>545705563Cloud computing is for retards
>>545706842Google search is built in AI.
>>545705563>buying overpriced computing hardware before it becomes 5 times cheaper in the upcoming bubble burst, instead of just paying $20 per month on a subscription and waiting for more favorable hardware prices>all of this to run weaker models and pay the $20 monthly anyways on electricity costs.don't really see the point.
>>545705563>run some Chinese shit that backdoors your pc bro, stick it to the man! The Alt-man, haha!The reason it’s a lost cause is to have beefy shit like opus 5.5 run locally you need like 2.5tb of vram and 4tb of ddr5. Yes fucking really. In the future as the models get more powerful the local running meme will be usable but far from cutting edge frontier shit.Thing is if people just want to run local to gen whatever porn or shitty slop movies they want or something, sure, existing consumer specs and open sourced models can get you there. But they’re not likely gonna get you solving millennium prize problems anytime soon or doing advanced scientific calculations. If all you want in life is to make as much porn as you want of SpongeBob anal voreing pregnant Spider-Man sure the local AI is a godsend for you. But most people just don’t need that.
Tried to watch his Kaleidescape video because I don't understand why people would pay $50k to stream movies but the way everyone in the video talked and their stupid little facial expressions was infuriating
>>545707069you quote and put -ai in the end
>>545706974I just got it a week ago still learning how to use it it's a ryzen ai max 128gb
>>545705865you think your shitty 1 gpu system can go toe to toe with an entire data center decompiling an entire program lmao
>>545706783>seems like it would only work for fairly specific jobsThe generic huggingface LLMs are just as all-round as chatGPT and such. Then there are domain specific ones that are focused towards shit like manage finances (don't), or decompiling software, and aren't good for anything else.But again, it's slower. Not because it's worse, per se, but because the datacenter neocloud companies are selling $100 worth of compute to the companies that serve the AI by running it on their machines for $5-10, to get people using the AI and capture the market. Expect some serious enshittification within the next year or two.
>>545706278Of course. You just need an attachmenthttps://www.lovense.co.uk/programmable-toys
>>545707243>this nigger thinks an entire data center is being utilized for consumer prompts Lol
>>545706576>some GB10 implementationI started with one of these, now I have two. The mistake a lot of people make is buying hardware first, when really you should start with the model. What are you looking to accomplish, beyond having local AI?>>545706671You can have usable AI for the price of a gaming computer today, the bar is much lower than it was just a few months ago.>>545707243Imagine being so out-of-touch that you think current models live only in VRAM.
>>545707098>instead of just paying $20 per month on a subscriptionThat the $20/month subscription price is a mirage, as it doesn't cover their costs. These companies are being subsidized. Eventually they'll have to raise the subscription prices or include ads in the models and enshittification will ensue.
>>545705563fuck off nigger
>>545705563This solution is better
>>545705563I would but people underestimate how much power it takes. Unless you're already rich, it's not worth it. Obviously power bill is one thing, but your hardware is precious, again unless you're rich. You do not want to run local AI on your machine and drain its lifespan.
>>545705563Racemixing tips?
>>545706836Hmm, I have an 8gig 3060 I might try with to see if it can handle anything bigger.
>>545705563imagine being an idiot and buying a computer when you can just pay $20 and rent one from the cloud for a couple hours every week. imagine thinking you are somehow owning a successful entrepreneur like Sam Altman by throwing away money on an overpriced home computer. fix your brown teeth commie
>>545707098ROI -1500%. Subscription should cost 500$, not 20$
>>545705797Google the Pinkertons. They still exist and do corpo wetwork.
>>545707473>people underestimate how much power it takesI'd argue people overestimate, unless you're talking franken-setups.
>>545707375I'm betting the bubble will burst before that happens, precisely because when they stop getting subsidized, the game of musical chairs stops, and their demand for compute won't be sustainable anymore.at worst price of hardware will stay the same or mildly increase, so I can always start self hosting when it becomes necessary.
>>545707697one more reason to get one.It's free money.not my concern that they're bleeding themselves dry.
>>545707697they are operating at a loss for the benefit of America so you can one day sit at home, enjoy life and have AI make a living for you while the communist democrats want to end AI, take america back to the stone age and send you to work in a gulag coalmine with your comrades. but at least they owned the chuds right
>>545705563He's a Kioxia fan, isn't he?
>>545706783It is unreal how retarded and corrupt the real estate market is in America, boomers have completely destroyed it and it’s no wonder no can afford a home. The entire appraisal system is the most kiked nonsense imaginable and realties have become literal pyramid scheme MLM insanity. Fuck realtors
>>545707473It's not bitcoin, anon. It doesn't run forever. It has task. It does the task. It's only active while you're using it, and it doesn't really have to consume any m ore power than playing a video game on ultra high graphics settings. I ran it for shits and giggles years back and it made no noticeable difference. And with AI distillation, they've only gotten more efficient since.
>>545707672And now it's all overhttps://youtu.be/4Dxdo1ErVY4?&t=258
>>545705865Local models can barely hit 70% accuracy and expert ai like coding and shit still need to access millions on websites that will take days for you to download and compute locally dor answers. Clause isnt just a neural matrix, its a few dozen programs mixing different ai and tools and precompiled datasets
>>545705563What would average joe need a local AI for?
>>545707366I have RAG running on my existing RTX 4070 with 12 GB VRAM and 96 GB RAM, with VL, reranker etc. and it works fine. However, I would like to run coder models, at least capable of talking Python, interactively. My current setup already allows that somewhat, but it's not exactly fast and I can't use RAG while I'm coding.I would also like to do some fine tuning on text to speech models, in order to be able to produce audiobooks in Slavic language.Checked.Dropping memeflag, got drunk yesterday.
>>545707999>It has task. It does the taskYes, and every single anon wants it to do infinity tasks. Whether coding or generating porn, that's more strain on your PC than playing a video game because AI requires maximum GPU usage, so your hardware is at risk of short term burn out. Only the most unoptimized piece of shit indie games would do that to your GPU.
>>545706203You know you can use small solar panels to keep tablets and smartphones charged indefinitely. That means you can access your survival books for years until your devices wear out.
>>545708164mined card are still running strong
>>545705563AI is not going away though. The military is using it. That means it will be funded by tax dollars, and federally protected. Anyone who tries to do something like they did with flock cameras will be treated as threats to national security
>>545707865I miss when this board wasn’t completely overrun with dumb migger boomers. You are genuinely retarded if you think any of that shit is going to happen. Here, take picrel, you’ll need it later
>>545707770>paying 20c/kwhyeesh I'm sorry anon
>>545707865>kikes>running at loss>for the good
>>545708122All that to code an on and off switch?
>>545707366shifting to memory is fucking loads slower than vram even if you were using 2tbs of ddr5, there is a reason why datacenters don't do it.
>>545708122>12 GB VRAM and 96 GB RAMThis is enough for Qwen Flash Next if you're okay using a vibe meme engine. That's mostly the meta now for "normal" hardware, an engine dedicated to the specific model you're using. The RAG doesn't matter at all because you can just point it at the new model you use.With one Spark I use this: https://github.com/blazux/qwen3.8-Flash-DGXWith two I use this: https://github.com/MiaAI-Lab/GLM-5.3-Flash-EXL3-2x-DGX-Sparks-TensorFoldQwen 3.8 variants are really good. The 27B is shockingly competent for its size and enough for coding if you have the VRAM for a decent quant. Flash Next takes some time to run because it reasons a lot, but its much more able to be a generalist for harness use. The only downside of the linked repo is that you can only manage one session at a time due to memory pressure. GLM 5.3 on two Sparks is the first local model that truly felt like a frontier replacement.>>545708338My cost is about half that, but I included delivery fees and taxes, then rounded up for a worst-case number. Unfortunately, delivery costs are often higher than generation costs where I am.
>>545707366>Imagine being so out-of-touch that you think current models live only in VRAM.It may as well only live in vram. If you're spilling over into ddr5 system ram or god forbid ddr4 you're probably quadrupling your iteration time at minimum. For heavy duty shit it becomes borderline not worth it if you can't keep everything on vram.
>>545708492Go look up what an ngram is. Then search how MoE models work. And we're not talking about datacenters, a single user local endpoint needs much less concurrency.
>>545705716Learn engrish
>>545708691Unified memory and >>545708694. This is why Sparks doubled in price.
So, what are you all using your fancy models for?
>>545705887>dont steal from corposRetarded goyslave
>>545708490Let the local models deal with the real world.>>545708517> RAGTrue, my embedder runs on CPU anyway.> Qwen3.8 competency I have to agree.> modelsThank you for the tips, I will test these.
>>545708758When that anon was talking about 1gpu setups I assumed we were talking about homegamer PC setups not standalone borderline enterprise grade shit. Yeah that's true of stuff like sparks but I don't think anyone thinks of that when the subject of personal hobbyist AIshit is being discussed anymore than people would assume you meant ASICs when the subject of btc or eth mining came up 5 or 10 years ago.
>>545707768nobody uses google anymore its 2034 anon
>>545705563Linus has some good ideas, too bad he's such an immense fag.
Is this guy even still relevant ?
>>545705865Claude is mostly great because of its proprietary in-house code. Unless some Claude leaks happen in the near future, the open-source models can't compete with it in recompiling so far.
Lmfao I rent pods and a b300 and put Qwen in it and it's like a fucking toddler compared to Claude so whatever piece of shit toaster this fucking faggot has at home won't run fuck all except a fancy chatbot
>>545709253I stopped watching Linus. He seems to be an insufferable asshole. Under paid his staff to the point they quit then went out and bought his own private jet.
>>545705628>A product so profitable, with such a vast market that they pay you to use it. It's literally unprofitable. It's negative profit. It's not even a bubble cause the wheel just turns while it slowly goes flat. Just another company hemorrhaging money lol.
>>545706203>GutenbergE-ink and unlimited sdcards, altho you can store the entire library in under 1TB. Battery lasts an age and easy to recharge with solar. Laptops in a SHTF scenario are ashtrays on a motorbike.
>>545709399luckiest salesman ever. he was in the right time at the right place
>>545705865I would wait next year or two. Opus 5.5 is great, and open models are nearing Opus 5.5's level. Hardware is going through a huge transformation right now. I can tell you that what is state of the art today will be inexpensive to run locally in a few years from now. Sadly it's the rule of Hardware two times slower than Software to catch up.
>>545709542https://www.youtube.com/watch?v=qOzbo3PYeww
>>545709399just like all communists he loves money he just doesn't want you to have any
>>545708835https://youtu.be/G7megUHUqFo
>>545709746really? bretty good
>>545709893https://youtu.be/kkwXVExCjg4This is also good as well. We have to aspire for something greater, the waifu and robotic future we were promised let's make it realized in the next decade.
>>545707117Lol, the mindset behind this post explains the state and cost of any public project in US perfectly.
>>545710141This
>>545710141>>545707117That's what ppl said about DrSbaitso and now he runs DARPA.
>>545706312The cost for them is training those ai models, not running them. Once trained, they barely have any costs. The problem is they don't recoup the training money before the next best model comes out.
>>545711514> once your datacenter has reached capacity it doesn't cost anything to run it anymore.
>>545706278You have to ask your mom for data to train it.
>>545705922Im okay with 1000 at year where do i sing it, i need someone to pay my electric bill
>>545709277has anyone gotten claude to dump its own code or is that locked up tight from day 0?
>>545705865you will get Claude like shit locally in 4-8years but in 8years the online claude will be miles away
>>545712351There was one huge leak in March, but that was long before it got good in recompiling.
>>545705563buy an ad faggot jew
>>545705563>That's just 50k $$$ goy
>>545712351Wdym the model? The CLI? The code app? The desktop app?Are you a retard?The model can't dump itself, the apps aren't anything special
>>545712479It was also just Claude code, which is a mediocre fork of vscode
>>545705865Got a strata node running (thanks fren) and it made good progress in recompiling shit i like. It's not Claude current day level but it's Claude something like a year ago and that's very finde for me for a local ran node with my own shit
>>545705563this will show you how inefficient LLMs really areit takes a local agent about 20 seconds to solve simple arithmetic
>>545705563i am appalled that people still watch this faggot, you have ruined my night, kill yourself.
>>545712562Local models are pretty fast. I wish I got an xtx instead of xt.
>>545709342Why would you use Qwen? That's not even close to the most advanced open weights model.
>>545705563I'm genuinely impressed this Jew went against a fellow JewI guess this Jew in particular was losing money and money for a Jew is more dear than another fellow JewBut then again the same applies to the CEO jew he liked money more than this fellow Jew and was making this fellow Jew lose money so this fellow Jew felt personally insulted so this Jew decided to strike back against the other JewBasically Jew on Jew violence >>545705628Where do I apply saar? But I want more money but having the electricity bill paid is not bad
>>545706203>I run my models to be ten steps ahead when the lights/internet go outThis is of utmost importance anons.Make sure you got some local (uncensored) models ready for when they start WW3 and shut down internet/power.People will die in troves when internet goes black. No one knows shit nowadays anymore, everything is just one google search/ai answer away
>>545705563i bought one of those big spinny niggers for 5 G's and use it to watch youtube vidz
>>545705865Yes, yes they're
>>545705563I sometimes ask the Ai to write my a story of hitomi tanakas tits in a hydraulic press.
>>545707243>you think your shitty 1 gpu system can go toe to toe with an entire data center decompiling an entire program lmaoYes, Dario
>>545708517Hello fellow New England nigger
>>545712444Lol noLocal AI will beat and surpass corpo AI, also corpo AI will lose access to their own ais as well, if only you knew what's coming glownigger
>>545712836LLMs are nice but you need fast ones plus external information sources (on your disk still but not baked in the weights knowledge I mean) so they can use rag as well to help you.But yeah I'm hoarding models. Should start downloading medical models as well.Also advice to you autists, NEVER EVER, FUCKING EVER, disclose or tell anyone you have local AI, normies will steal it from you, subhumans cannot be trusted under any circumstances
Btw the best part of the video was that he is struggling putting together PC hardware correctly (The fan cable exiting through the atx opening, kek)You see that he didn't do PC building for like 10+ years
i walked into radioshack and sauntered over to the computer section and says to the guy, gimme one of them big lightning speed niggers so i can play AI videos like my grandson
>>545705563Local models such and are slow. For now AI companies are heavily subsidizing use (or have focused on efficiency like DeepSeek so can potentially serve at a profit)
>>545705915It's not gonna make a difference if they regulate it. They will just move the data centers to other countries.
>>545705563WTF is this? He just builds an overpriced PC designed for suckers than the AI doesn't run.Did Sam Altman put this little race mixer up to this?
>>545705865do you think an AI works independently from the hardware on your computer? like some kind of spirit?
>>545705563>be zogmerican>"openai">is closed source>be chyna>open source ai>zogmerica calls to ban it
>>545705947>Grok, watch this video for me and summarize it in a sentence.
