6,575
edits
Meta Platforms: Difference between revisions
→Product & Usage
| Line 67: | Line 67: | ||
== Product & Usage == | == Product & Usage == | ||
June 13, 2023: Meta releases I-JEPA, a human-like AI model that potentially has the ability to analyze and finish uncompleted images accurately compared to other AI models. This model uses background knowledge it has about the world to finish the images<ref name=":1">https://finance.yahoo.com/news/meta-releases-human-ai-image-171319516.html</ref><ref name=":1" />. | |||
May 18, 2023: Meta has build custom silicon chips to help with AI and Video-processing tasks<ref>https://about.fb.com/news/2023/05/metas-infrastructure-for-ai/</ref>. Its vice president of infrastructure Alexis Bjorlin said the improved performance resulting from these chips will justify the huge costs associated with building them<ref name=":0">https://www.cnbc.com/2023/05/18/meta-pulls-the-curtain-back-on-its-ai-chips-for-the-first-time.html</ref>. For instance, Meta Scalable Video Processor (MSVP) chip is used to process and transmit videos while cutting down on energy requirements as Bjorlin noted “there was nothing commercially available” that could be used to transmit and process 4 billion videos a day as efficiently as Meta wanted<ref name=":0" />. The other processor is the Meta Training and Inference Accelerator (MTIA), which is intended to handle "inference" and is aimed at powering Meta's ad recommendation algorithms<ref name=":0" />. Meta added that it has also completed the final phase of its supercomputer known as Research SuperCluster (RSC), which will train Meta's LLaMA language model. These new hardware will eventually be used to operate tasks associated with Metaverse<ref name=":0" />. | May 18, 2023: Meta has build custom silicon chips to help with AI and Video-processing tasks<ref>https://about.fb.com/news/2023/05/metas-infrastructure-for-ai/</ref>. Its vice president of infrastructure Alexis Bjorlin said the improved performance resulting from these chips will justify the huge costs associated with building them<ref name=":0">https://www.cnbc.com/2023/05/18/meta-pulls-the-curtain-back-on-its-ai-chips-for-the-first-time.html</ref>. For instance, Meta Scalable Video Processor (MSVP) chip is used to process and transmit videos while cutting down on energy requirements as Bjorlin noted “there was nothing commercially available” that could be used to transmit and process 4 billion videos a day as efficiently as Meta wanted<ref name=":0" />. The other processor is the Meta Training and Inference Accelerator (MTIA), which is intended to handle "inference" and is aimed at powering Meta's ad recommendation algorithms<ref name=":0" />. Meta added that it has also completed the final phase of its supercomputer known as Research SuperCluster (RSC), which will train Meta's LLaMA language model. These new hardware will eventually be used to operate tasks associated with Metaverse<ref name=":0" />. | ||