Meta Unveils New AI Supercomputer (wsj.com) 48
An anonymous reader quotes a report from The Wall Street Journal: Meta said Monday that its research team built a new artificial intelligence supercomputer that the company maintains will soon be the fastest in the world. The supercomputer, the AI Research SuperCluster, was the result of nearly two years of work, often conducted remotely during the height of the pandemic, and led by the Facebook parent's AI and infrastructure teams. Several hundred people, including researchers from partners Nvidia, Penguin Computing and Pure Storage, were involved in the project, the company said.
Meta, which announced the news in a blog post Monday, said its research team currently is using the supercomputer to train AI models in natural-language processing and computer vision for research. The aim is to boost capabilities to one day train models with more than a trillion parameters on data sets as large as an exabyte, which is roughly equivalent to 36,000 years of high-quality video. "The experiences we're building for the metaverse require enormous compute powerand RSC will enable new AI models that can learn from trillions of examples, understand hundreds of languages, and more," Meta CEO Mark Zuckerberg said in a statement provided to The Wall Street Journal. Meta's AI supercomputer houses 6,080 Nvidia graphics-processing units, putting it fifth among the fastest supercomputers in the world, according to Meta.
By mid-summer, when the AI Research SuperCluster is fully built, it will house some 16,000 GPUs, becoming the fastest AI supercomputer in the world, Meta said. The company declined to comment on the location of the facility or the cost. [...] Eventually the supercomputer will help Meta's researchers build AI models that can work across hundreds of languages, analyze text, images and video together and develop augmented reality tools, the company said. The technology also will help Meta more easily identify harmful content and will aim to help Meta researchers develop artificial-intelligence models that think like the human brain and support rich, multidimensional experiences in the metaverse. "In the metaverse, it's one hundred percent of the time, a 3-D multi-sensorial experience, and you need to create artificial-intelligence agents in that environment that are relevant to you," said Jerome Pesenti, vice president of AI at Meta.
Meta, which announced the news in a blog post Monday, said its research team currently is using the supercomputer to train AI models in natural-language processing and computer vision for research. The aim is to boost capabilities to one day train models with more than a trillion parameters on data sets as large as an exabyte, which is roughly equivalent to 36,000 years of high-quality video. "The experiences we're building for the metaverse require enormous compute powerand RSC will enable new AI models that can learn from trillions of examples, understand hundreds of languages, and more," Meta CEO Mark Zuckerberg said in a statement provided to The Wall Street Journal. Meta's AI supercomputer houses 6,080 Nvidia graphics-processing units, putting it fifth among the fastest supercomputers in the world, according to Meta.
By mid-summer, when the AI Research SuperCluster is fully built, it will house some 16,000 GPUs, becoming the fastest AI supercomputer in the world, Meta said. The company declined to comment on the location of the facility or the cost. [...] Eventually the supercomputer will help Meta's researchers build AI models that can work across hundreds of languages, analyze text, images and video together and develop augmented reality tools, the company said. The technology also will help Meta more easily identify harmful content and will aim to help Meta researchers develop artificial-intelligence models that think like the human brain and support rich, multidimensional experiences in the metaverse. "In the metaverse, it's one hundred percent of the time, a 3-D multi-sensorial experience, and you need to create artificial-intelligence agents in that environment that are relevant to you," said Jerome Pesenti, vice president of AI at Meta.
Who is this meta you're talking about? (Score:5, Insightful)
Re: (Score:3)
Glad to see I am not the only one annoyed by this name change.
Lots of corporate execs collectively came up with a bad idea.
anyway...
Do you think anything about this sounds familiar?
Obviously, AI has been in the headlines a lot.
Supercomputers plus AI seems like it might be a dangerous combination.
Large scale operations like that present many opportunities for corruption to sneak in.
i think the distinction between "user" and "lab rat" can start to become quite arbitrary, as
vast numbers of servers are deployed
What? (Score:2)
Too subtle?
Flushing a brand name (Score:2)
Coca-cola should change their name to X. Because X is such a cool letter.
Re: (Score:2)
Re: (Score:2)
Maybe Elon Musk can sell his domain to them.
"Meta" stands for.. (Score:5, Insightful)
"Metastasize"
Re: (Score:3, Interesting)
metastasize
intransitive verb
1. To be transmitted or transferred by metastasis.
2. To be changed or transformed, especially dangerously.
3. To spread, especially destructively.
sounds perfect for facebook (Score:1)
Re: (Score:2)
And it's learning also on private data. (Score:3)
Just be aware that this supercomputer AI is also using private data for learning.
Re: (Score:2)
It's such a waste of resources too. Developing things like language models are a dead end. They might make speech recognition a bit better, although it already appears pretty much perfect to me, but it will never lead to a general purpose AI, or strong AI.
All it does is suck up vast amounts of electricity and valuable GPUs, to deliver something that we already have, presumably because Facebook doesn't want to licence it from other companies.
Re: (Score:2)
And is due to become self-aware in 60..59..58..
A waste of resources (Score:5, Funny)
Unfortunately ... (Score:5, Funny)
It can only process metadata ... :-)
Re: (Score:3)
It won't be the military that builds skynet (Score:3)
It'll be FAANG
MAANG (Score:1)
More stupidity, faster! (Score:2, Interesting)
Artificial Ignorance is truly making great strides. If they continue like this the machines will eventually be even more stupid that the average human. Then we are in real trouble!
Re: (Score:2)
Artificial Ignorance is truly making great strides.
Given how it's used to evaluate if access should be blocked (ex: Google accounts, etc.), for unwarranted reasons, without human intervention (and no way to appeal), I instead call it "Artificial Incompetence".
Re: (Score:2)
At least for Google, I definitely agree.
Re: (Score:2)
I don't understand (Score:2)
The aim is to boost capabilities to one day train models with more than a trillion parameters on data sets as large as an exabyte, which is roughly equivalent to 36,000 years of high-quality video.
But how many Libraries of Congress is that? or for the Metric folk how many station wagons of reel to reel tape?
Re: (Score:2)
Does that help?
This bears repeating (Score:5, Interesting)
It's FACEBOOK, not meta.
Meta is the name Facebook has adopted for its tax-dodging holding company, and to fool people into thinking they're something other than privacy-invading data brokers.
Seriously, on a site like SN, could we at least tack on the real name of this company each time Meta is mentioned? Like Google / Alphabet? I expect better from this site than to help perpetuate this farce.
Re: (Score:1)
The word "meta" has been ruined. Meta-data ... owned by Facebook. Metaverse? Owned by facebook. The world just needs to dump the word "meta". Or, since it's a generic term with pre-existing art, Facebook shouldn't own the word, and other companies should rename themselves to "Meta" as well. There are twelve cities named Meta already! https://geotargit.com/called.p... [geotargit.com]
Re: (Score:2)
Hard to believe (Score:2)
Really? This? From the company that can't figure out how to identify and eradicate terrorism and hate speech? I seriously doubt they can build the "metaverse" when they have no clue about the real world.
It might be the fastest, and yet... (Score:3)
It still won't tell you that the Apollo moon landings were real, that the Covid vaccines are good for you, and Trump lost the presidential election by a landslide...
Re: (Score:2)
Lots of stupid calcs fast is not intelligence (Score:1)
We will never have actual AI while AI means doing lots of stupid calculcations very fast.
Re: (Score:2)
Skynet. . . (Score:2)
Apparently the real-life Skynet was start as a Facebook AI. And here we thought it would be military.
Re: (Score:2)
Re: Skynet. . . (Score:2)
A better comparison would be the AI programs in the William Gibson novels. They are usually created by corporations, and because they are often huge secretive projects that outlive their creators, they start to behave in unpredictable ways or continue to function after they were thought to be dismantled.
The scary part is that the nefarious things these AIs will do will be baked into the programming. Facebook wants to mine your personal data and manipulate you. That means they are programming AIs to learn to
It wasn't the Crypto Miners After All (Score:1)
Now we know what really caused the GPU shortage!
And can we please call our future "Oasis"-verse something *NOT* with "Meta" in it? We really don't want evilmeta attributed with creation or ownership. It's like Microsoft renamed itself to "computer" or "operating system". Can the people sue evilmeta for appropriating a perfectly useful adjective?
Does anyone honestly believe that 'Social Media' (Score:2)
What exactly will the endgame be? I think we are really going to f**k the human race here.
Just my $.02
"High quality video? (Score:2)
Too bad (Score:2)
the computer will be used for evil and to find up better ways to serve up ads and keep mankind enslaved to the machines of social media.
Missing something (Score:2)
AI research is clearly missing something - when the human brain can do "intelligence" with far smaller data sets, and can do it consuming only about 20 watts, what are we missing that we require on the order of ten thousand heavy-duty computing devices and "thousands of years" of training data, and still in many way be more cognitively limited than a toddler?
Re: (Score:2)
But our brains are doing a lot more than just running data through a giant neural net with weights and getting some output. The brain is performing different types of operations on d
How to locate the AI Research SuperCluster (Score:2)
video volume on PC browser (Score:2)
36,000 year training set is a red flag (Score:2)