There is a strong belief that moderating posts from billions of users can somewhat be a difficult task but Facebook believes strongly that artificial intelligence can bring some form of ease. And that is why Facebook’s parent company is taking steps daily towards achieving one goal, which is building its first supercomputer.
Recently Meta announced its new AI Research SuperCluster (RSC), it is a supercomputer that was created to carry out AI research projects. The said supercomputer has been in creation for the past two years, the social media giant believes that with the new supercomputers they may just eventually be able to develop powerful AI software with sophisticated abilities. Something potentially useful for spotting no so pleasant posts such as hate speech
A software engineer, Shubho Sengupta at Meta while commenting about the supercomputer to a news source said “With RSC, we can more quickly train [AI] models that use multi-modal signals — for example, to understand the context of a post, including language, images, and tone.”
In recent years of AI research, Supercomputers which comprises various features that are interconnected processors clumped into what are known as nodes have become increasingly popular and very powerful. Investigations into things like unfamiliar proteins have been aided with The US Department of Energy’s Summit which is reportedly the fastest supercomputer in the US and the second in the world. A number of large companies playing in the tech space have supercomputers for their own and used for various purposes. Tech giants such as Microsoft, Nvidia to name a few.
In a recent blog post by Meta, the company revealed that said that its supercomputer currently includes 6,080 GPUs divided among 760 nodes. With figures this huge, potentially its supercomputer is one of the most powerful the world currently has. Sengupta reveals that a supercomputer of this current size is at par with the fifth most powerful computer in the world, which is the Perlmutter supercomputer present at the National Energy Research Scientific Computing Center. Meta has hinted that when it said computer is eventually completed later this year, it will have 16,000 GPUs. The company also expects it will be capable of nearly five exaflops of computing performance — that’s 5 quintillion operations per second. According to Sengupta, “When fully built out, the cluster will almost triple in size, which will, to our knowledge, make it the fastest AI supercomputer in the world.”
Meta has said that with the supercomputer its researchers will be able to “seamlessly analyze text, images, and video together” as well as come up with new augmented reality tools. Also to be used for training of large models related to natural-language processing and computer vision. Over time, the company hopes that with the supercomputer’s completion its vision of “building entirely new AI systems” that can ultimately perform computationally difficult tasks is achieved. Early tests have revealed that the supercomputer has the capability to train large language models at a rate that is three times faster than the system it currently uses. What this simply means is that a supercomputer will train in 3 weeks what its existing system would take nine weeks to train.
Eventually, Meta has revealed that technologies needed to build the “metaverse” would be led and aided by the supercomputer. A virtual realm in which people can carry out various activities like play, interact, shop get an education, visit places via digital avatars. While the company rebranded, we were told that the metaverse was its near future ambition, though there is still a long way to go we can for a fact that snippets of that vision have started to be realized.