News Elon Musk's xAI raises $6 billion to build even more powerful AI supercomputers — Nvidia, AMD contribute to funding round

[Admin]

xAI, which is set to build a supercomputer with 200,000 Nvidia GPUs, raises $6 billion.

Elon Musk's xAI raises $6 billion to build even more powerful AI supercomputers — Nvidia, AMD contribute to funding round : Read more

 

[Konomi]

They'd have no problem hitting funding goals if Elon just says he wants to play Crysis. I'm sure everyone would chip in to get him a GPU..

 

[AkroZ]

A valuation of $50 billions when you generate only $100 millions annualy, it's over valued.

 

[Sleepy_Hollowed]

Without considering the scam artist this guy has been shown to be (space x succeeding despite him aside), this is insanity.

AI is a bubble, but his stuff is the worst of all, and even if it’s hardware only, Facebook has been doing custom hardware for at least a decade.

 

[hotaru251]

Kind of glad he is in TX now and won't put all this useless power consumption on the nations grid and can just stress TX's special grid more (sorry texans)

Hope someone can figure out how much money every "ai" company spent whe nthe fade/bubble pops. Wanna see which company ends up biggest lost $.

 

[thisisaname]

Elon Musk spending other people's money not a new thing.

 

[SomeoneElse23]

I simply don't understand the bad feelings people have towards Elon.

Nor how they don't hesitate to say bad things about him every chance they get.

Merry Christmas all.

 

[bit_user]

Kind of glad he is in TX now and won't put all this useless power consumption on the nations grid and can just stress TX's special grid more

So far, Colossus seems to be based in Tennessee:

https://www.tomshardware.com/tech-i...enabling-all-100-000-gpus-to-run-concurrently
​

If he wants to continue scaling, it obviously will need more locations. I've heard electricity can get very expensive, during Texas' summer heat waves, so that might be a reason for him not to build there.

 

[bit_user]

Major investors in this funding round include Nvidia, AMD

It feels a little like extortion to strong-arm your hardware into investing. I'm not saying that's what he did, but you can imagine the conversation:

"Hey guys, I'm looking for investment in my AI venture and sure would appreciate it if you could chip in. That way, maybe I could afford to buy (more) of your hardware (wink, wink)."
​

With his recent surge in political clout, that'd make it all the more complicated for someone to turn him down.

 

[waltc3]

Elon Musk spending other people's money not a new thing.

They all spend other people's money, all of them, without exception. Musk is merely spending a lot less of other people's money than they are...😉 I like that myself, because in tech, like in politics, he who spends the most money frequently loses the race. Quality matters far more than quantity. You have to have a certain quantity to attempt a business, but if the quality isn't there, then fogedaboudit...😉

This whole enchilada seems more like an adolescent e-peen contest every day. A lot of that, I think, is because of the way it's publicized in the general and tech press. It's all so simple-minded these days. It's not about computers that think, reason, and judge, because those computers do not exist and never will. "AI" seems to me to be essentially a new face--a new spin--on search engine tech. It's what the copyright lawyers are hashing out right now, and Google is hoping to avoid with the FTC.

Forget Skynet and complex computers that "become self-aware" one day. That's all investor dope designed to milk the pockets of the unwashed masses. What are the base requirements for becoming an AI investor? Four things strike me as essential:

1) You really enjoyed the Terminator movies and Skynet "makes perfect sense".
2) You see investing in Skynet, "AI," as "getting in on the ground floor of a great investment, like you did with Walmart."
3) You have an adequate amount of disposable income you can safely risk.
4) Greed rolls off of you in waves, and you regard it as an essential food source.

Yes, I know, I'm cynical, but I'll wager I'm not alone...😉

 

[thisisaname]

They all spend other people's money, all of them, without exception. Musk is merely spending a lot less of other people's money than they are...😉 I like that myself, because in tech, like in politics, he who spends the most money frequently loses the race. Quality matters far more than quantity. You have to have a certain quantity to attempt a business, but if the quality isn't there, then fogedaboudit...😉

This whole enchilada seems more like an adolescent e-peen contest every day. A lot of that, I think, is because of the way it's publicized in the general and tech press. It's all so simple-minded these days. It's not about computers that think, reason, and judge, because those computers do not exist and never will. "AI" seems to me to be essentially a new face--a new spin--on search engine tech. It's what the copyright lawyers are hashing out right now, and Google is hoping to avoid with the FTC.

Forget Skynet and complex computers that "become self-aware" one day. That's all investor dope designed to milk the pockets of the unwashed masses. What are the base requirements for becoming an AI investor? Four things strike me as essential:

1) You really enjoyed the Terminator movies and Skynet "makes perfect sense".
2) You see investing in Skynet, "AI," as "getting in on the ground floor of a great investment, like you did with Walmart."
3) You have an adequate amount of disposable income you can safely risk.
4) Greed rolls off of you in waves, and you regard it as an essential food source.

Yes, I know, I'm cynical, but I'll wager I'm not alone...😉

The other side of point 3 is your broke and your putting it all on AI to save you, hight risk high reward strategy.

 

[kep55]

xAI, which is set to build a supercomputer with 200,000 Nvidia GPUs, raises $6 billion.

Elon Musk's xAI raises $6 billion to build even more powerful AI supercomputers — Nvidia, AMD contribute to funding round : Read more

Come one, come all. Get yerself some of Dr Elon's AI Snake Oil. It's guaranteed to answer all your questions, build all your code, and run all business processes. All one of these little bottles will cost you is a measly dollar. Per nanosecond. With little to no direct usability and consistant user interface. Each bottle comes complete with an exclusive no-safety-cap container and a voracious appetite for everything, especially other people's data and electricity.

 

[JRStern]

It feels a little like extortion to strong-arm your hardware into investing.

It's a bit of a con game by NVidia, they "invest" $1B, turn around and use it to buy their own products at retail, pocket the 50% markup, but deduct $1b when calculating gains, if any. Same deal if their "investment" is simply made in product rather than cash.

 

[bit_user]

Musk claims that xAI benefits from data drawn from X.

Doesn't this tell us all we really need to know? So long as his training data is garbage, the GIGO principle applies. Doesn't matter how many GPUs he trains it on.

Maybe he'll figure this out, sooner or later, but I doubt he'll admit it.

 

[SomeoneElse23]

This whole enchilada seems more like an adolescent e-peen contest every day. A lot of that, I think, is because of the way it's publicized in the general and tech press. It's all so simple-minded these days. It's not about computers that think, reason, and judge, because those computers do not exist and never will. "AI" seems to me to be essentially a new face--a new spin--on search engine tech. It's what the copyright lawyers are hashing out right now, and Google is hoping to avoid with the FTC.

I agree with you. AI, as it's currently hyped, is nothing more than Search Engine 2.0.

It's a big improvement over existing search engines, but still terribly unreliable.

And certain not intelligent.

 

[jp7189]

Flux has nothing to do with xAI or Musk, but is the product of the very awesome people at Black Forest Labs.

Aurora is the xAI image generator, which has been "unnamed" to be part of Grok.

 

[jp7189]

Doesn't this tell us all we really need to know? So long as his training data is garbage, the GIGO principle applies. Doesn't matter how many GPUs he trains it on.

Maybe he'll figure this out, sooner or later, but I doubt he'll admit it.

Nah.. you just need a lot of horsepower to sort through it all for the data classification part. Train on crap data a model that recognizes good v crap, and even though it sucks at it.. its still 1% better than the method before. Reclassify your sources 1% better, train a new model, rinse and repeat. That's what a whole lot of GPUs can do for you.

 

[bit_user]

Nah.. you just need a lot of horsepower to sort through it all for the data classification part. Train on crap data a model that recognizes good v crap, and even though it sucks at it.. its still 1% better than the method before. Reclassify your sources 1% better, train a new model, rinse and repeat. That's what a whole lot of GPUs can do for you.

The fundamental problem with Twitter is the limited amount of really good data on there, as well as the kind of "good" data it has. Most tweets are short and snippy, not just because of the traditional 140-character limit, but also just how the platform is designed. Its ranking algorithm (which governs the likelihood of tweets showing up in peoples' feeds) rewards attention-grabbing and inflammatory posts which draw lots of reactions. Those responses are then very reactionary and probably rather low-value. It's not conducive to thoughtful, longer-form posts and articles.

I just think it's not such a strategic asset, the way he's talking about it. Something like Reddit would be much more valuable. Twitter data would be great for training a toll bot, though.

 

[jp7189]

I would imagine xAI gets access to the "un-algorithm-ed" data... which would be an advantage no know else has. I'm not knowledgeable enough to know how much good data is on Twitter, but there sure is a lot of it. Also, it must be said "good" is pretty subjective. If they expect grok to help with python code.. Twitter might not be the best data source. If they want to build a chat bot and later a human interactive robot that can be engaging and witty and helpful.. then maybe Twitter is a good start.

 

[bit_user]

I would imagine xAI gets access to the "un-algorithm-ed" data... which would be an advantage no know else has.

The main problem with Twitter's algorithm is that it trains users to write a certain kind of post. The posts favored by the algorithm get lots of likes, so users will write more posts of that sort and you get a lot fewer of the sort of posts that don't "drive engagement". In other words, their entire dataset is biased. The filter just amplifies the effect of what's already a highly reaction-driven medium.

If they want to build a chat bot and later a human interactive robot that can be engaging and witty and helpful.. then maybe Twitter is a good start.

I think the best training sets for this sort of thing are either transcripts from real chat sessions with a human online support tech that were graded highly, or perhaps some of the better Question & Answer threads from Stack Overflow-type sites or even this one. Those give you a "Best Answer" and individual posts often have an additional score. Also, the individual participants often have a score of some sort, which can be used to further hone in on who the community seems to consider experts. That can help your algorithm or humans curators select which are best for inclusion into the training set.