Meta Plans to Launch Generative AI Model by 2026 to Support Reels and More
-
Meta is fully committed to generative AI technology, planning to introduce a new AI model capable of supporting video content by 2026. According to Facebook President Tom Alison's extensive discussion at the Morgan Stanley Technology Conference in San Francisco, the company's technology roadmap will focus on developing an AI model to drive the video domain, including Reels and longer-form content.
Currently, Meta uses separate AI models for each of its products, such as Reels, Groups, and Home Feed, each with its unique recommendation system. However, Alison stated that the future goal is to shift from providing unique models for each product to a unified ecosystem while enhancing user engagement with recommendations.
Alison explained: "For example, we are pushing a project to use a single model to support our entire video ecosystem. Then, can we also have our Feed recommendation products served by this model?" He revealed that Reels accounts for one-third of Facebook's total viewing time, meaning two-thirds of the time is based on the existing user configurations of the social media company. However, to achieve this ambitious plan, the company needs to address numerous technical and user-related issues. Allison pointed out: "All of this requires significant hardware investment and planning. Therefore, apart from this, we are actually reconfiguring data centers and exploring how to connect more GPUs together."
To achieve this goal, Meta stated that it is working to scale up its models with appropriate hardware and data support, partly due to the need to analyze interactions from billions of users. Currently, Facebook has over 3 billion active users, which means a massive amount of computational processing is required. Additionally, according to Allison, factors such as advertising and user behavior also need to be considered.
This plan is just one part of Meta's broader AI goals. The company also aims to enhance its chatbot assistant to work in tandem with improved recommendation systems and increase responsiveness. Allison provided a use case: if a user is recommended content related to Taylor Swift, they could use Meta's assistant to find the date of her next concert without waiting for comments or leaving the platform to search. In another case regarding Groups, users can rely on Meta AI to answer baking-related questions faster than human commenters. Ultimately, the AI can also send users relevant previous posts related to their questions, and users can interact with the AI.