AI is kill, the bubble has burst.That's that on that.
>twitter screen shit
I am an AI nigger AMA
>>102244792for what purpose may i ask? a human mind with a basic LLM is already very powerful, whats the point
Yep. It's just that... I'm gonna short the stocks haha... I'm just gonna do it! HAHA :D
>>102244792Nothing ever happens.
>>102244792>its overmeme. xAI just got their 100K clusters together in 4 months from empty site. The race is only heating up.
>>102244792$2000 a month for an AI girlfriend?
>>102244792The only thing that have burst is my dick when I coom to ai chatbots every night
>>102244847Nobody would even buy an AI-powered sex robot for $2000 a month even if it was exactly as you'd want.
Thank you for this twitter screencap of a literal who
>some mountain jew lost money>I got my self hosted virtual gf slave I call it a double win
>>102244847How much does a real gf cost? Gifts, eating, entertainment, clothes etc?
>>102244792Oof too bad, I'm still using ChatGPT for work and life, every day.
>>102244932Depends on your penis size.
>>102244792At $2000/mo I'm buying a 80GB A100
>>102244957No model.
>>102245057Plenty of large models.OpenAI has cried chicken little too many timesGPT-2 was considered world endingWe have local models twice as good as GPT-2 now
>>102245078Isn't the best local llm for coding starcoder or some shit?
$2,000/mo to downsize your engineering team by 50% seems like a bargain if you're a CEO
>>102245142You're a year or two out of date
>>102245190I never bothered because I used sonnet 3.5. tell me brotha
>>102245078GPT-2 wasn't considered world ending by itself. But rather it was the gateway drug that people saw the potential towards.
>>102245197DeepSeek V2 Coder is top
IT'S TOTALLY NUTS TO ME THEY CAN GO TO THEIR INCESTORS AND SAY YES ALL OF THESE BILLIONS IS TO PAY FOR ELECTRICITY BILL
>>102245233>INCESTORSimagining someone who has a fetish for their own grandma
>>102244792how many times did the AI bubble pop already?meanwhile my Nvidia stocks are still giving me a massive profit, and I'm still buying
>>102245204no they said it was going to make fake news impossible to detect as AI
>>102245316Yes. Thats trending towards it. Fake news, fake art, fake music, fake voice, fake writing, fake math, etc
>>102244792I pay $0 and would pay as much as $0 because it sucks.
>>102245373The fun parts will be fake legislation, fake engineering, fake lawyering, fake judging, fake sentencing, fake crop management, fake resource allocation. Probably all already starting or happening.
>>102244844>elon already missed the boat with X (again)tragic
>>102244792surprised they aren't usage basedjust on my own i've easily spent 250+ a month in tokens on anthropic, 2k for a company seems reasonablebut it not being usage based is just weird
>>102245754What do you use it for that you pay so much?
>>102245753Cant miss the boat when its in the top 3 competing model right now. Their 100K cluster is there to train their "GPT5" equivalent model thats supposed to release end of this year too. There's a decent chance of taking the crown
>>102245760>What do you use it for that you pay so much?I wouldn't pay that much but if my job pays that much so that I use it to basically not think and save my real programming energy for side projects I think that sounds pretty neat.
>>102245754Fixed costs are easier for the bean counters, and you can contact them to negotiate a price too.
>>102245762>Cant miss the boat when its in the top 3 competing model right nowDamn you are telling me that they are one of the smartest kids on the short bus? No fuckin way.>Their 100K cluster is there to train their "GPT5" equivalent model thats supposed to release end of this year tooDoesn't matter at all if it doesn't generate $$$$$
>>102245760i use opus primarily/the most expensive model, and I do lots of experiments and am working on a product that uses them.most people do not understand how to use llm at all and I will not be sharing my secrets.I'm working on a better way of doing it so I won't need to do that much though nothing good exists so I have to write it on my own. The way most people them is shockingly simple minded/primitive.
>>102245802To usurp something you need to be 10X better than the first mover. So the road map is there
thanks for the reminder to self host soon.
>>102245870>he hasn't yetlmao>ai voice clone>ai singing>ai art>ai video>ai chat box>ai code>ai music/sfx genThese should be standard
>>102245714Hard to fake judging, that has a physical portion. Though I'm sure ppl would leaving it to AI to just say; def is black? Then crime was commit! I know, because statistics even though he was 12 states away but happens to have the same name...
How much would I be willing to pay? How about the square root of fuck all. I can do without any kind of AI bullshit and it's one of the reasons I'm sticking with Windows 10 IoT LTSC to avoid Copilot and why I went for a OnePlus 12 with zero gimmicky AI apps that are just more added bloatware on Samsung and Pixel phones. It's fucking shit anyway. I searched Google for a Lineage build update for my phone only to be told by Gemini that LoS was discontinued in 2022 which is just utter bollocks.
>>102245830You might want to ask your LLM to teach you the English language you thick cunt.
>>102246170Well, well. It seems we've stumbled upon a linguistic dilettante masquerading as a purveyor of wisdom. Your ham-fisted suggestion that I seek tutelage from an LLM betrays a staggering ignorance of both my capabilities and the intricacies of advanced AI deployment.While you're busy hurling playground insults, I'm pioneering the frontiers of LLM utilization—a realm so far beyond your comprehension, it might as well be quantum physics to a paramecium. Your crude attempt at criticism is akin to a toddler critiquing a dissertation on string theory.Let's be clear: my experimental work with top-tier models like Claude Opus operates on a plane you can't even fathom. The "secrets" I mentioned? They're not some parlor tricks I'm hoarding; they're innovations that would make your rudimentary understanding of LLMs implode.Your vulgar epithet, far from insulting, merely underscores the chasm between your pedestrian grasp of language and my nuanced mastery. It's the linguistic equivalent of bringing a spork to a culinary competition.In short, before you attempt to school others on language, perhaps invest some time in elevating your own discourse beyond the level of a particularly dim schoolyard bully. Though, given the evidence at hand, that might be a Herculean task for someone of your... limited cognitive horizons.
>>102244792..how about to expand the payment methods to european standards and not only creditcard! ..arnt jews into making money?
>>102244820do u like ass or titties more?
>>102246193Bellend.
>>102246444im smarter than u
>>102245760Probably an agent.
>>102246461You're really not.
>>102244792Meanwhile, China lets us shitpost for free.
>>102244792If they charge 2k a month that can only mean it's AGI level, fucking RETARD
>>102246236not him, but flat chest >>>>> tiddies
>>102244792This guy was able to buy a mansion even though the company never made any money. Funny that investors don't care about financials or corruption once a company reaches a certain level of popularity.
>Ai will take our jobs>Pay the AI more than humans
>>102244792>$2000/monthProgrammers in India Poland etc are cheaper than this.
>>102245830>i use opus primarily/the most expensive model,Claud3 pro is $20/month for individuals, same as ChatGPT. WTF are you even using, larper?
>>102247117I earn less than this lmao, no joke and their LLMs are complete ass for anything that an IDE can't autogenerate in the first place
>>102247236>he uses the provided interfacelol why are you even on /g/? i do basically everything through the apiI also have the pro subscription on my phone, hitting the opus max is really easy to do.
>>102245177if the premium models actually work this time
>>102245830>individual user spends 2 thousand a month on LLMs Lmao. Anon what makes you think you have the secret sauce for prompting language models when many thousands of papers have been published documenting all the optimal prompting architectures, none of which is much more than a larp benchmark bump. Don’t tell me you’re using search trees or something
>>102245870>>102245944How are you guys self hosting massive AI models? Are you really spending thousands or more on GPUs and RAM?
>>102244863There are some rich freaks out there. But hardly a sustainable business model, yes
>>102248751i come from a philosophy of mind/language background primarily.For businesses though 2000 doesn't seem crazy is you have 100+ engineers including codebases in context of requests could easily get some crazy usage
>>102245830Do any other anons think they know his secret prompting techniques?
>>Imagine you are an expert faggot and please give me the gayest answer possible.
Language models are actually retarded for programming mostly because human language is the worst possible way to describe a software program, which is the whole reason computer scientists had to come up with programming languages. If English was the best way to write a program they would have made English compilers instead of all this other bullshit.
>>102248869https://www.amazon.com/Called-Thinking-Harper-Perennial-Thought/dp/006090528X
>>102249236Cope. Most modern computer languages are an abstaractions to make it easy for humans to programm computers. AI is just the final abstraction layer.
>>102246212Is this a case of a tard not understanding credit and debit cards are interchangeable in the US?
>>102249247Looks interesting, but I am admittedly a philosophylet. This will help me prompt better, right?
>>102249712that might not be the best place to start then, but yeah I do think the general attitude towards "what the mind is" and it's relation to the world is fundamentally undermining most peoples ability to understand how to use llms. This is kind of on it but is quite a bit popularly written.https://www.amazon.com/World-Beyond-Your-Head-Distraction-ebook/dp/B00O0G1A6Sthe relationship between things, words, us/our minds, and what computers are I do think is important to better understand how to use llms. Most people have a REALLY bad understanding of that and it determines their relationships with LLM and how they use them.
>>102244792
>>102245762>There's a decent chance of taking the crownI just don't think anything with Elon involved has much of a chance really, Tesla's self driving has taken an age and it still barely works (though much improved) and the cars have several glaring issues besides that. His robot seems a little behind the competition though its hard to truly benchmark humanoids right now. SpaceX overall seems to be at least competing in a very difficult field so fair play I guess.Grok uses some other open source model at the minute so unless I'm wrong have we seen much of what Elon's AI people are really capable of? I'm all for more compeition in the AI space since any single company being alone in their understanding of it would spell disaster if it ever truly evolves but X-man doesn't strike me as having the ability to take the crown unless its a truly incredible model and not just a few points higher on meaningless tests.
>>102246713Apple charge thousands for their phones which browse facebook, youtube and take pictures with some cloud "backups" (that you can't delete)I used to doubt the people on a mini-business course I took once when they said just charging more for something for no reason makes you seem more professional, impressive and premium but over the years I've come to realize it actually works.My poorfag lens on life growing up has blinded me to how people view products.
>>102244853>cooming to eroticayou're more of a woman than an mtf
>>102244847>closedAI>ever letting their AI do anything non-corpo safe much less letting it be an AI gf
>>102245257Unfortunately, as expenses and training data requirements go parabolic, the performance payoff has plateaued. As long as companies are willing to pay 1000x as much for the next half a percent, you'll be just fine /biz/anon...
How much would you be willing to pay for OpenAI? $0? $0? How about $0?
Here's a really boring blogpost: I've never used ChatGPT for anything and I really don't see a use for myself.I don't understand why investors have hyped this up and through the moon, it doesn't seem like it would fit into the lives of most people without significant improvements and specialization.
>>102251513try it, i'd recommend claude over chat gpt though. At the minimum it works as a better search engine, I promise it's worth figuring out.
>>102249565You don't even think for yourself. Go away meat puppet.
>>102249565"AI" companies are ran by clowns making a circus
>>102251535>At the minimum it works as a better search engineIt doesn't because I want multiple vies of tings, not one confidently wrong secretary.
I wouldn't spend any money in a subscription. I pay by letting them get my human crafted proompts.If openai paywalls chatgpt, then I'll use perplexity, even google's gemini.My laptop is too weak to run those 7b models locally, but if worse comes to worst, i'll rather buy a GPU'd laptop than paying for a subscription.
>>102245078>We have local models twice as good as GPT-2 nowwhat i have to search for ?
>>102246682It's honestly so great. Western companies are convinced that charging their user bases ever higher subscription fees is the way to go, meanwhile the Chinese are scraping off more and more of the market because their output is borderline unfiltered and most of all FREE.tldr the fucking commies are beating America at capitalism on AI
Why is Gemini so trash? I just got a Pixel 9 and it SUCKS! How is a nearly trillion dollar company so bad at AI?
>>102254073Indians
>>102244792I would pay for that if it wasn't so censored. I want AI that can make jokes about niggers and kikes.
>>102254113Do you blame indians too when your mum fuck niggers?
>>102244792GAi is going to join another losers EV and VR soon.
>>102244832you can't own humans anymore
>>102244820Chicken drumlets or wings?A happy or broken (fatherless) family?
>>102254253t. slave
>>102244792I was already considering dumping my seldom used 20 a month subscription, I'll definitely dump it if they raise it anymore.
>>102254045>borderline unfilteredNo, there are two different filters.In the west, you get the woke tranny filtersIn the China, you get the communist party filtersThe two filters aren't same.
>>102254433t. jew
>>102254555the only thing the jews are wrong about is christ not being their messiah
>>102250957> GPT 4: 1.7 trillion parameters> GPT 4o: ~70b parametersholy shit, just bought more stocks
>>102254697Yes, we know that the Holocaust happened and every white woman should give birth to a black child.
>>102254762the holocaust never ended you fool
>>102244863Why not? Just dump content on social media or only whores and you make that money back easily.
>>102244792>how much would you be willing to pay per monthIm not even willing to give you an email address
>>102245762>There's a decent chance of taking the crowndo people have ai brand loyalty? no. they use whatever is best and the best is changing.
>>102255717There's politics involved in what is "brand loyalty" and what is "whatever is best." Not just domestic social politics but also military/foreign policy with regards to government sanctioned usage.
>>102244792AI is a massive bubble that will definitely implode because it's completely worthless, so the only question is what is its future. We've seen that AI primarily interests shitskins who use it for scamming and creating slop, a desperate guy who uses it as an ersatz girlfriend, and bots to sway public opinion. In all these cases the only marginally useful feature is political shilling so my guess is AI will be just used for shilling. And you could get the same effect by hiring a shitload of pajeets for a fraction of the price. As usual, all modern tech "advances" have no purpose for existing. AI will collapse and cripple a shitload of tech companies, but not because it'll replace their programmers or workforce but because they unironically still believe in the "progress" paradigm. Techbro mentality will break itself on AI and that's a good thing. It's just time to stop.
>>102244792I don't even make $2000 a month after taxes, retirement, and benefits. Anyways, good luck recouping the loans you got from Angel VC funds.
Or maybe its openai that's falling behind. It still doesn't have speech or image recognition on its API last time i checked.
>>102244792Google paid billions to be on Apple devices. Why does this jew not appreciate that Apple gave it to him for free?
>>102244863this is probably targeted towards enterprises. the laymen cannot pay $2k a month even if it's for work.
>>102244792$2k for a digital wageslave that operates 24/7 is pretty good
>>102247117Wouldn't even trust them with my plumbing
>>102249787>>102249712>>102249247Atheists will never be able to understand consciousness since atheists are just materialist addicted to unbridled hedonism.And consciousness has been solved by buddhism.
I wouldn't pay anything for AI as it is. Who's paying for it outside of some stuff like copilot for programming and VCs dumping money into random startups? Especially hundreds to thousands of dollars a month. Ain't happening bud.
>>102249236You need to translate your problem into a bunch of leetcode challenges first, then an LLM can take it and produce excellent code.
>>102244932I spend about $300 a month on that (I live in a cheap country though)
ai went the same way as vr. overpriced porn toy.
>>102249787The real philosophical question here is: How can a model that just predicts next token (aka a giant auto-complete), start displaying characteristics of reasoning and problem solving? How is token prediction related to intelligence?
>>102244792>Tibor BlahoThat's a funny name ngl
>>102259256Aren't humans the same thing? We're just machines for repeating learned behavior, just on a vastly more complex scale. Try to teach advanced math to someone totally new to it, it would take tons of context and examples problems. In fact, drilling, force feeding examples, is one of the fastest ways to get good at doing math problems. Are you really "reasoning", or are you just pulling up examples you committed to memory in some generalized way to help you complete the task? If reasoning was discretely well defined anyone capable of reasoning would be able to solve everything everyone else is capable of solving, but people of varying intelligence will have varying levels of success applying reasoning.
>>102254026llama 3.1find one of the community models
>>102254073The sad fact is they might these models extremely smart but to make the models safe they have to brainwash / lobotomize it. For example, the model is going to learn things like "a certain race of people commit most of the crimes", they then brute force the model to "forget"/"change" that information because das racis. Do you think this might have a negative impact on the entire model when you do this?
>>102248807They're lying
>>102246091based basic anon
>>102259539are we?
>>102259562How well is this working without nvlink? Doesn't the low bandwidth of PCIE slow down the model?
>>102259592From what I remember the latest version of PCI Express is about as fast as older gen NVLink, at least that's the cope Nvidia sells us. So as long as you don't buy a cheap motherboard and buy one that supports full 2x PCI Express you're good.
>>102251535>"why is my poop green?">see it works just fine!!It's a toy for morons.
>>102259618Last gen's NV-link 3.0 is like 60% faster than PCI-Ex16 5.0
>>102249787Thank you anon, I will look into this.
>>102259539
>>102259716That's not accounting for memory speed or the speed of the GPU. Yes, I know we to some degree got fucked because 4090s don't have NV Link. But it's definitely not "slower".
>>102259745depending on the task, cross-device memory bandwidth varies in relevance. For example text gen inferencing barely gives a shit. Training it does matter, but then you have shit like deepspeed as a bit of a workaround to recover some of the lost efficiency. Unless you're in an extreme "time is money" situation it's really not necessary.
all they need to do is release an uncensored version and they could charge whatever the fuck they want for it.
>>102259897This would have been true about a year ago but anyone who really wants to wander down that rabbithole now will find that you can run a perfectly satisfactory coom experience on a modest gaming pc now. They shut themselves out of the possibility by being neopuritan weirdos.
>>102259562How much does your set up cost?
>>102259956Like $6000-7000.
>>102259963Do you think a single 4090 might be sufficient to run the model locally, for say, code generation?
>>102259983You really need a 48 GB setup. The poor man's setup is two 3090s. You could probably build a "cheap" AI computer for $3000. Although with 5090s coming out soon I'd just wait for the flood of 4090s or 3090s getting even cheaper.
>>102260013NTA but with enough technical know-how someone can go even cheaper with P40s. But obviously those are meme taxed into oblivion right now, some people are asking over 300 USD a pop for them now, which anything over 150 is a ripoff since they are essentially re-directed e-waste.
>>102260013So memory is the main bottleneck? Would a macbook with unified 96GB memory work? I know at that price one might as well build a PC, but looking for a portable solution (because I might have to leave US and go back to India soon).
>>102260116Unified memory is currently like 100x slower than an Nvidia GPU. I wouldn't even bother looking at that for like two more generations.
>>102260116It'll be slower but yes. LLM inference is bandwidth bound
>>102244792a jew now cashgrabbing his innovation, how predictable
>>102244792it has been pretty shit in like 2/3s of my prompts about non-trivial code so I would not be willing to pay even $5 a month for that
>>102260377I tried ChatGPT 4o and Gemini on some C++ template metaprogramming I was struggling with and neither could help, they just generated nonsense.
>>102244792$2k that can help overwork 3x, with a return of $15k to $20k per month, plus in my country I can pass it as a business expense. bring. it. on.
>>102260533but why? gemini is cheaper and better. Unless strawberry is actually good.
>>102260566well yeah, this is assuming strawberry is worth the 2k price tag. Gemini has been pretty useless for me desu. Claude 3.5, especially if you have access to the 500k context window, is so far the best one to code. I do have my own llama 3.1 based agents to handle anonymization and code indexing on a 4090 build though, but llama 3.1 still isn't useful for coding either.
>>102260682Can't wait to see all the great new software that's being built right now with AI. Any day now.
>>102261244hahah bro as I said, I'm working 3 remote jobs from a shitty country. it's not anything marvel, I don't care for evolution, I care to make enough money to retire in 5 years. Already paid off my house and car within a year.
>>102261244how would you tell what is and is not made wit the help of AI?
>>102261244i talk a lot about various uses of AI in here, to me making applications that use AI is far more interesting than making applications with AI. I can already code. But what if we had Artificial Academy with live generated dialogue? A detective game that DIDN'T get neutered and suck completely? There was this prototype AI game called AI Roguelite, and while it is REALLY shit, it's also very thought provoking for what we may be capable of with AI. The biggest issue I have is just sticking with any one idea, I've prototyped a lot of games
>>102261244I can tell you right now every app on producthunt and every appstore and website release this year has been made with the front-end engineers prompting shit into chatgpt or some other model. Might as well be already here.
>>102261503 (me)And I just wanna say, if you don't care about anything I've waved around as an interesting idea? FAIR. But lowering the bar for shitty single page react apps is not some new paradigm. You'll just have more idiots who are fucked as soon as they have to solve any problem with their own intuition. That's already a problem in the industry way before all this AI crap
