Facebook parent company Meta disclosed new information about its artificial intelligence (AI) projects in several blog posts on May 18.
Diving into AI technologies
Meta built a first-generation silicon chip for its AI models under the name MTIA (Meta Training and Inference Accelerator) in 2020.
The company said it designed this chip to serve as a complement to GPUs. It said that it used this inference chip to improve recommendations and is now working on new versions of the chip with higher specifications.
Meta additionally said it created a new data center design that is AI-optimized and the data will be used in AI applications, including inference and training. It will also complement the company’s video transcoding chip, Meta Scalable Video Processor (MSVP).
Finally, Meta said it is building a 16,000 GPU supercomputer for AI research and called the machine “one of the fastest AI supercomputers in the world.”
AI efforts remain largely private
Meta has not revealed much information about its AI technology in detail. However, it has announced details of AI products aimed at the public in the past.
So far this year, Meta and its executives have announced plans for AI-powered advertising and have discussed the use of AI in content discovery and user support.
The firm’s AI language model, LLaMA, also leaked to the public in March against the company’s intentions, giving the open-source community access to the technology.
The post Meta reveals new details about in-house AI chips appeared first on CryptoSlate.