Facebook parent companyMetaInvesting heavily in artificial intelligence technology as part of its technology road map to 2026, aimed at developing newer technologiesArtificial Intelligence Recommendation ModelRecommend all of its video products, including its TikTok-like Reels short video service and Facebook FeedTraditional long videos in.
Meta executive Tom Alison revealed that the company currently typically uses separate recommendation models for each product such as Reels, Groups, and the main feed, but will develop a unified large-scale model across product services in the future.
FirstThe stage is to migrate the existing recommendation system from traditional chips toGPUImproved overall performance. The second phase, which began to build this new “model architecture” last year, has been tested on Reels, adding 81 TP3T to 101 TP3T for the duration of Reels' viewing for Facebook core applications。
Alison stated that Meta was appalled by the ability of such large artificial intelligence models to “many data and tasks”. As a result, in its third phase, Meta will endeavour to validate and promote this technology and use a single model to recommend services for the entire video ecosystem and even Feed。
“These recommendations will be more attractive, relevant and responsive.” Alison said that if you saw what was interesting in Reels, you could see similar recommendations when you returned to Feed。
Meta has purchased a large number of GPUs to support broader AI work, such as developing digital assistants. Some plans include integrating complex chat tools into feeds, allowing users to ask AI assistants questions about a tweet topic; or allowing group members to ask AI assistants questions related to the topic.
Alison said, “We have the opportunity to place the generated AI in a multi-person consumption environment”