Not long after its rebranding and announcements of creating a complete metaverse, Meta (formerly Facebook) has presented the AI Research SuperCluster — the AI supercomputer.
The supercomputer in question has been a part of the company for almost two years now. However, Meta now hopes that this supercomputer could become a far superior AI software than anything we’ve seen so far. With just a bit of training, of course.
Software engineers believe that they can use it to train AI models to recognize hate speech better than before. They will use multi-modal signaling to help the AI learn how to distinguish between different contexts, images, and even tone of videos.
Meta’s supercomputer currently stands at 6,080 GPUs which makes it one of the most powerful in the world. The company has already begun building new units and they expect that the supercomputer will be complete by the end of 2022.
If their estimates turn out to be correct, the AI Research SuperCluster will have 16,000 GPUs which will make it able to perform 5 quintillions (5,000,000,000,000,000,000) operations in a single second. After this happens, it will by far be the most powerful supercomputer in existence.
The company believes this is a crucial step forward since it’ll lead to the creation of technologies that will become an integral part of their Metaverse. It will consist of a range of AI-trained avatars, interconnected realms, and technologies able to mimic and interpret natural language perfectly.