>Samsung is making more money selling ram to data centers than with anything else>AI companies valued over a trillion dollars>all products are available for free>radiologists don't need chatGPT they can just run AI locally>you can make images locally and don't need data centers>you don't need AI companies and they're giving it all for free because you don't need them>AI hasn't produced anything meaningful to humanity>CEOs of irrelevant companies adding "AI" to their scripts because they know it makes stock go up, like in the dotcom boom
>>61561589Its a good enough grift to justify bringing the US electrical grid into the 21st centuryAI is generally useful and our power infrastructure is dogshit, its a win win and isnt going to kill the stock market or whatever, probably
>>61561589>radiologists don't need chatGPT they can just run AI locallyimage recognition had been at a point where it was useful to radiologists for many years before chatgpt was even shat out and yet there are still radiologists. generative AI is completely irrelevant for that.
>>61561589>>you don't need AI companies and they're giving it all for freeYes that was the original idea, it was called "OpenAI" because it was supposed to be open source. LLMs were cool and interesting when that was still the premise. They then gaslit everyone into believing that LLMs would somehow morph into AGI and destroy the planet and therefore we needed to be first to develop them. What we have now is not an "AI apocalypse", but rather a horrific exercise in human retardation left unchecked
>>61561589Not to mention picrel is gayJust a reminder, nothing wrong with it
>>61561664>think it's fake>it's true>moreover, this is what these guys actually look like in profile
>>61561589The problem of AI is the "Good enough" problem. At what point is there a model that is "Good enough" for people to run locally for free without big tech. At that point dumping exponential amounts of energy for a marginal gain in predictive power is going to seem unprofitable. The problem is the data centers will have been built, the power stations will have been built, but the will to expand will die. There will be no sense continuing to use the current gen chips to improve the model 0.000001% and no will to build bigger better data centers. These data centers will go dark and the power delivery will be diverted to something more useful.
>>61561589The money is made by kicking you jeets out of your job.
>>61561801that is not a problem at all. LLMs both have historic data aggregated as well as perform current online searches. historic data is like a big chunk of the currently available internet/archive. e.g. more than Terrabyte in size. Even if Moore's law still was true (it is not anymore), we would need several decades before it would be feasible for a small company to run it's own complete LLM. The other "AI" applications like medical analysis discussed in this thread are not LLM/AI at all, just machine learning (buzzword for pattern recognition). Not to mention, the corporate fucks have illegally scraped the internet for the mentioned training data. You as a small fry try to scrape it yourself (so not to rely on DeepSeek or some other chinese dataset) and you better be prepared to be fucked in the ass by the government.
>>61561739that photo is extremely edited. the guy on the left has baby legs compared to his upper body
>>61561589The only way to run an LLM is in a datacentre, whether you use yours or rent somebody else's. And in both cases it neess a TON of RAM and a ton of chips.
>>61561615>Its a good enough grift to justify bringing the US electrical grid into the 21st centuryBy the time this is over the Electrical Grid won't even resemble 20th century America.You don't understand just how devestating this bubble popping would be. Electrical grid contracts would collapse, Power plant contracts would collapse, all of their contractors and subcontracts would fold overnight, and this isn't even getting into the domino effect this would have on tech and the various tech and hardware companies that now have contracts with these AI companies.If even one of them, OpenAI, Microsoft, Google, default on any of the loans they've taken out to make the next trillion dollar industry, it's unironically over for the US as a first world nation. This is why the FED is so interested in printing money to keep them solvent.
>>61562623You can run deepseek locally
>>61561801The trick is knowing that we already passed that point.>>61561873>You as a small fry try to scrape it yourself (so not to rely on DeepSeek or some other chinese dataset) and you better be prepared to be fucked in the ass by the government.My man, you can get their weights for free. You don't need to scrape historical data yourself, the big companies like Meta already do it for you and let you use it however you please.https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct
>>61562623This is absolutely false. You can run llms on local machines, the new ai 39x chips from amd and m series chips from apple have enough memory bandwidth that you dont even need a fancy nvidia card to run local llms .
>>61562644How is the fed giving them money? Aren’t they just buying bonds from banks to the tune of 40 billion a month? It’s just regular old QE. Also, as far as I understand, OpenAI isn’t in danger from loans but rather their commitments.
It won’t pop. Governments have been convinced they are all in a race to build a machine god and the first one gets to rule the world. Bailouts and backstops will happen, environmental concerns for buildouts will be smacked down. Strap in for the ride anon.
>>61561615>our power infrastructure is dogshitSays WHO?
>>61563736Interesting times
>>61561617Yeah but now it’ll kiss your ass and tell you what a great doctor/ radiologist you are while gently correcting your mistakes.
>>61563917NTAI’ve heard from someone who supposedly knows, that overall our power infrastructure is pretty robust but that obtaining transformers and turbines have a big backlog
>>61563690You can run them, yeah. You need more compute than is affordable without investors/benefactors to train a high end model though.
>>61564153So, consumers can run them, but they can't make them (high end models) and the goal is to keep advancing by training more advanced ones, and that's where the need for large compute and energy use comes from. Although running them locally does require enough compute that Google Colab quickly banned them because that many people running them locally at once was an overload for even google
>>61562665not the full size deepseek, aka deepshit
AI is going nowhere, these jews need to build their AI moshiach/antichrist/beast ala Deus Ex, they'll start as many world wars and bleed humanity as much as necessary in their mania to construct the ultimate idol.
>>61563959It gets worse when already humanity's quality of life is getting more stressed because of increase in speed to develop it. >its as if AI already has control is and currently solidifying its power
>>61561589Are there any good open source AI models trained for financial research? Or anything investing related?
>>61561589Its just part of there plan to buy up everything. And try to get to big to fail for a bailout. Why else would they promise all these huge deals with other companies when they'll never be able to raise the money
>>61562665>>61563690You can, for ERP.I mean any business for doing anything remotely serious. The smallest use case. Still requires a datacentre with tons of chips and tons of GDDR, clustering and spares and yes, they will need clustering and spares if you don't want your business to go down.
>>61565425>Still requires a datacentrewhat can you use latest chagpt for, though? if you buy the enterprise datacenter chatgpt option.can it actually like do something useful?
>>61565287>>its as if AI already has control is and currently solidifying its powerNever thought about it that way
>>61565331The general meta is to use a shitty small model locally, but pay for chatgpt 12 or whatever for a few months and just run whatever specific use case you have through chatgpt but fine tune the local cheap model.After a few months the local model will handle most of the use cases and you cut the chatgpt API calls.
>>61564170>the goal is to keep advancing by training more advanced onesWith what data? All data available has already been consumed.
>>61565287T. Read that one xeet from strawberryfag
Other than the formation of Roko's Basilisk in real-time, where is the money behind these literal trillions of USD of investments in AI supposed to come from? There is no product being developed through AI, there is no increase in the efficiency of industrial, agricultural, scientific, or cultural yield that can make even a fraction of what's been put into it back, and there is nothing being generated through prompts that won't immediately be invalidated by the existence of someone's knowledge of said prompt that can generate the exact same model as well. The race to AGI isn't something with a clean forecast for a timeframe (half because they keep lobotomizing AI for Israeli interests, half because development is being hampered by lowest-bidder hires making the bulk of the scientific workforce so companies can save money in the process) and even with the development of AGI, it will still be years before it produces anything of concrete value beyond its own existence as a theoretical infinite value engine. I could trust that a large portion of this is billionaire fossils being scammed out of their fortunes by some of the craftiest techno-jews in existence and simply being played at all levels because they see some numbers going up today, but there has to be a point where they realize that financial autofellatio can only sustain itself if there's something else to be brought in. What exactly is the sales pitch for another $100,000,000,000 investment into a data farm, chip supply, and research budget into this process?
>>61561589>AI hasn't produced anything meaningful to humanity
