AI IS NOT A BUBBLE!STOP COMPARING IT TO THE DOT COM BUBBLE!
SAARS, may we have a better source other than twitter's anus?
>>107052346Why does he need to tell this?
>>107052346Well it's much worse because during the 2000s the economy was doing fine overall.This is going to be more like 1914 / 1929
Great now he officially hexed it
he doesnt want it to pop before hes out and have it be his legacy
>>107052346Cry about it in your BUBBLE.
>>107052346AI isn't going anywhere
>>107052570its already everywhereshatbots are shatbots doewhat theyre actually capable of is kinda written on the tinthe bubble is that people counted on chatbots to compute the answer to everythingit wont disappear but oh lawd is there a market correction incomingand its gonna be glorious
he's right, it's much, much worse
>>107052713>speculation sink>the fucktards who pump the economy into being based on a whole heap of nothing, one of the major causes for inflation are gonna get the cookie jar lid slammed on their fingersfrom a macro-economical standpoint that bubble may actually be a good thing
>>107052346>Technology is viable, but costs are sky high while revenue are minimalIt’s actually a worse version of the dotcom bubble
>>107052673i think chatbots will largely go the way of the dodo because theyre way too inefficient, but smaller ML projects will remain as they have for 15 yearsalways found it funny when people said engineers are going to be replaced, ignoring that no one is going to let an AI sign off on a project like that for decades, solidworks has had ML and automated matrix math for 10 years
>>107052346>>>/biz/
>>107052346he's rightAI bubble is not the same as .com bubbleProbably worse
HAHA BUBBEL MAKE BOOM
>>107052876chatbots are a good interface.you can build/train em to emit labels which then get fed into various frameworksand ai engineering already existsits just that people who use it dont share and its a serious technological investment to build onehttps://en.wikipedia.org/wiki/Model-based_designhttps://www.youtube.com/watch?v=VdG4gUTowXcpicrel has been designed by ai.not a chatbot, obvioulsythe pipeline goes like this:engineer gives a set of constraints: dimensions, materials, (manufacturing methods in a generalized engi ai)the ai then iterates bw generating a model conform to the constraints and testing it, attempting to remove material each iterationso in a way it is gen ai, it even uses something akin to a generation/evolution algorythm, but without the statistical aspect of ML. its fully symbolic
>>107052346That Powell fella seems on the up and up! Better believe what he says.
I'm 25% cash but the market keeps booming. Is diversification for chumps?
>>107053062this is not how booming looks like
>>107052958NOOOOOOOOOOO THE US GOVERNMENT'S $1,000,000,000,000 DONATION TO SAM ALTMAN!!!
>>107052758No they aren't. We're all going to pay to bail them out.
>>107053072My 1 year returns are 17%. The market is booming.
>>107053230i dont think its realisticwere not talking several billies hereits trillionsat least a couple, bc applel, ngreedia are vastly over-valuated + openai & co if they go full bankrupt, thats something like 3-4 trillionsand theats before counting the derivativeswhich get speculated on too, each layer of repackaging gradually adding a layer of speculative valueits gonna be glorious
>>107052758When has a bubble ever been a good thing? The AI bubble bursting is going to kickstart the Great Depression 2.0
>>107053260>bases prediction on past performance instead of looking for signals that allow to predict the futuredumb money. but enjoy yourself, somebody has to buy the bags after all
>>107053297>When has a bubble ever been a good thing?id argue- every single timethe great depression hapenned because a whole nation didnt do their due diligence if im not mistakenthe definition of a bubble is speculative value getting annihilatedits shadow value that skews the valuation of a fiat currency into inflationi think its a necessary evil + a wake up call
>>107053304Not making predictions, just talking about right now. If you want to make predictions: I'm 25% cash and regretting it what are you investing in to profit off what's coming?
>>107053346im supposed to do 5-30 mins trades on crypto, and ive been on a sabatical year since 2 years now, so im not actually informed about the current state of the market, i dont follow the numbers, and i play in the junior league anywaysbuuuuutgold is always a refuge in a period of turmoiland it will probably spike once the bubble will popbeware of paper gold though, because i know theres a swindle with gold papers (etf/etc?)once the demand will explode people will actually look into the actual availability of gold and all the swindles may come to light.theres a scenario, and i think its moderately likely where gold explodes twofold: first bc suddenly everyone wants to hedge against the bubble, then once the swindles come to light, and solid gold value goes through the roof for a moment
>>107053304>predict the futureKEK
Gold: probably safe during crashReal estate: in a bubble already, would avoidBonds: treading water, will survive crashCash (high yield): Same as bondsThere has to be something better. VIX?
>>107053346>>107053382dont go all in on gold eitherdiversify. gold will overheat at points, and there will be correctionsdiversify away the fluctuationsmaybe invest in some hard value like housing, idk, landidk man, do due diligencebut gold might be a thing to look into
>>107052346>>107052363Once OpenAI shutdown then there is no customer support and the skeleton crew who is still working at support will be overwhelmed or crashes the system because 500 calls to one center is not a good idea.Emails for Openai will be sent to real people who may take years to finish their job.There will be no more updates until corporate deals with angry employees who were fired or need to be retrained. New hires to replace the old will slow down updates.So called artist will try to use local machines AI but soon realize how much money they are losing to electricity. They will never achieve the same results as openai because the real open source models suck.AI is a bubble, it will hurt only USA who over dependent on it. This is your American economy problem, not everyone else.
>>107052346How could it be a bubble when demand is far, far outstripping supply?
>>107053426simple rule of thumb:you take the netto yearly revenues of a companyand you divide by its valuationthis gives you how many years would it take to recover your investmentif you were to become the sole owner of a company at its current valuationthen you compare to the rest of the industry/financial landscapewhen a company yields less than a savings accountyeah, its a bubble
Frankly, if you take away my ability to use AI, my life will be severely impeded. Everyone I encounter also talks about how they use AI to improve their daily life. It changed the world and is an essential utility now like electricity and water.I'm not going back to google searchesI'm not going back to writing boilerplate codeI'm not going back to authoring PDFs from abstract dataI'm not going back to photoshop or fucking gimp for memesWe're just not doing that shit anymore
>>107052346He's right. It's much worse.
>>107053443Listen, very carefully: change, or you will fail miserably.
>>107053458i thought i already said im a high-ish frequency trader, so i have a big tolerance to riskand im quite versed in risk mitigation techniques.and even im gonna say: ai is overheated as fuck. limp dick / 10, would not buy atm
>>107053458>>107053469contsmart money invested in ngreedia & co when covefe started and people got gibbesnow that train has sailed, buying into it now would be buying high to sell low
>>107053469lel if anything it's underbought because of fud like thisyou don't understand tech trends
>>107053535>looks at failure rate of vc capitallolleyeah i dont understand the hypeand thats a good thing. i dont intend to change that
>>107053535>sar please buy the AI very smart
>>107052346>everyone gets into AWS cloud services until live is jacked up and everyone needs to scale back and find cheaper alternatives >this definitely won’t happen with all the AI trash we’ve invested in for 2 years thoJob security, simple as.
>>107053552I look forward to taking your bag in 3 years lmao
>>107052519
>>107053625ill diversify your mom away in exchange for goods and services
>>107053449Time to let go of your toy and face the future.
>>107052346Yeah, it isn't the same; it is multiple orders of magnitude worse.
>>107053753not gl its gonna kinda suck if/when chud gpt diesits a powerful search engine, you can give it complex queries, you can partially bypass the paywalling of scientific documentsi kinda like it
>>107053449You can just run qwen3 locally. It only needs like 4GB of RAM. There's no way this goes away now.
>>107052346Based. Total luddite death
What can you even do with gigantic GPU farms running LLMs if it turns out there's no sustainable commercial demand for LLMs?
>>107053881We're about to find out.
>>107053881crypto?
>>107053881>>107053889you sell emwe might see a tsunami of heavily used enterprise gpus, but at a bargain pricecryptobros would have a field day, thats for onesmaybe users are gonna massively buy em for local models purposesmaybe theres some agent-class ais being developed to become opensourced, that would be basedfukken, who wouldnt like to have a jarvis with voice recognition and generative abilities running locally
>>107053914i didnt really think about that issuebecause using an llm to do that is just so much more convenientand i dont need the actual texts either. in my usecase it boils down to finding a figure, or getting a high level overviewan llm with browsing capabilities is sufficient for that
>>107052514well fuck.
>>107053928>agent-class aisLol, lmao.
>>107052346
>>107053980???llm into labelsthe labels are actually commandsslap a speech to text on top of it for mechanime maximization purposesthis could actually increase productivity.and would be cool as fuck
Don't worry, they just need another few trillion dollars and then they'll finally be profitable!
>>107053980>>107054009aaand with some generational capacities bc its still an llmits gonna happenand its gonna get opensourced at some pointbut i dont want to wait 20 years for that to happen :/
>>107054019they hit the wallits completely delusional to attempt to scale up at this pointits hard maths.consider two tokens, in the context of, well, context size: a and bnow consider a unit of operations, an abstraction of the computing needed to build a two token correlation mapthis gives us 2 token context size => a->b,1 unit of computation.now lets consider 3 tokens, a, b and cno we have a->b, a->c, b->c3 units of computation, 3x the compute for a 50% increase in context sizethe computational needs grow pyramidally with the increase in token size.this is the problemand beyond a certain point, its just completely hopeless
>>107053928>you sell emidk they might shred them right before their loans get called. You're overestimating public corporations.
>>107053842Maybe in 10 years? I only use the most advanced models, and the free shit can't compete.
>>107054201why would they do that?if i were the lender id be pissed that my investment is a total loss
>>107054134trump just needs to print goyillion dollars and the growth will look linear again
>>107053881Pointless question when the growth of AI has been faster than the growth of the Internet itself.WE ARE NOTGOING BACK
>>107054201>>107054253contin 2008 they didnt destroy the housesthe flooded the market with em
>>107054254thats not exactly how that happens.theres extra steps to it and not all of the printed money goes into ai
>>107054280spend on ai grows exponentially while results are lower and lower, they need to start printing in hyper mode or the whole economy collapses
>>107054305just few more quadrilly bros
>>107053928You're probably gonna see companies just buying hardware, dismantle if and sell the components to MSI so they can make regular consumer GPUs
>>107054305nahi think you underestimate the size of the us economy+ letting the economy collapse is macroeconomically the right thing to do+ that would be a big win for the protestant gang bc that would decimate the jewish gang's assets
>>107053449You are a subhuman who outsourced his ability to think and create and thus has no reason to exist.
>>107054336its refurbished thenit would be a major scandal if they tried to sell that at "new" pricei dont think its worth the falloutif they just refurbished the cards and sell them at a decent price...that would be niceor maybe something in the middle: they sell refurbished+downsized hardware to poor markets like india
AI winter has already started. We might have another breakthrough in 10 or 20 years.
>>107054352inb4 kike altman shills: AGI race is too big to fail, everyone needs to bail them out
>>107053297Post-industrial, financiarized economies need bubbles to keep the economy growing.
>>107054352The Jewish gang are going to sell all their paper assets to goys just before the cataclysm hits, anon. Like they always do.
>>107054439they already did, your retirement money is already in it goy, time for bailout
>>107053449Why do you need to do any of those things? And boilerplate code has nothing to do with AI. Most IDEs have been able to automatically write boilerplate code for like the past decade.
>>107054378i think its too big to bail outits not only the ai companies that would need a bailoutbut everyone who didnt sell in timeand some products are derived from ai investments, which will drop the trust factor of many investors which will mean massive withdrawals and speculation down for many investment fundsi dont think everything is gonna happen at once thoughits hard to imagine a scenario where one day, everyone sells everything ai relatedi think well see an inflection point, where people who buy into ai will no longer outweigh people selling, which will prolly look like a plateau for a moment, then a crashmost ai companies arent even public yetso theres no curve to follow, properly
>>107052346It's not like the dotcom bubble because the internet provided something truly revolutionary.AI provides a financially suicidal autocomplete that takes billions of times more processing power to enable retards to generate buggy unfinished code that they can't fix without more prompts
>>107054439they need bailouts every 5 minutes.i think youre greatly overestimating them.people who got fucked over in 2008 didnt do basic due diligencebailouts ensued.they influence immigration policies to maintain a predictable market growth. bc theyre retards who can only hodl.yeah, they have money etc, but give me a familial fortune to start with, plus insider trading information, and youre gonna see im a second warren buffet
>>107054366Please use a chisel and hammer to build the Empire State Building. Unless you've outsourced your ability to think and create to machinery.
>>107054499Oh, so you're just really ignorant.
>>107054134In terms of brute force compute, memory bandwidth, etc, sure. We’re reaching the point of diminishing returns, but there is a long way to go as far as optimizations are concerned. Hardware architecture will continue to evolve as well.
>>107054570ye ofc but its not happening fast enough and thats how big tech ends up considering the acquisition of nuclear fukken reactorsmultistage turbokek
>>107054570>>107054600aaaaan even in terms of optimization theres only so much you can do when the root problem is exponential-likewhat is needed here is a paradigm changewhich means ETA: unknowncould be tomorrow, could be 5 years from now
>>107052346>>>107033485Oof. Bubble so tight they're making photoshop's clone free to shove AI into it.
>>107053260You literally described a bubble.
>>107054375At least there's still lots of ingenious work to do when it comes to integrating LLMs as language interfaces to expert systems and gaming. And it will stay a nerd thing because normies can't be trusted with not killing themselves by accident due to some hallucination.
Big tech investing in nuclear power is a positive thing.>>107054634Sure, AI/ML won’t become O(n) any time soon, but we’re still scratching the surface on optimizations. Algorithms today do a ton of unneeded computation, the hard part is figuring out what shortcuts you can actually take. Look at how fast transformers and diffusers are improving. You’re right that we need a paradigm shift, and I think some kind of hardware architectural breakthrough is what will happen. The real answer is finding a better way to avoid time complexity than “just add more tensor cores, bro”
