Meta spends billions of dollars on semiconductor chips as part of its aspirations in the field of artificial intelligence, with these efforts placing a special focus on its video-centric platform.
NVIDIA’s image processing units have become the foundational chips relied upon by researchers in artificial intelligence to develop large language models used to power conversational robots like ChatGPT and other creative AI models.
A company executive mentioned that the significant investment in artificial intelligence covers the establishment of a sophisticated AI system specifically designed to manage video recommendation systems for Facebook across all its affiliated platforms.
Tom Allison, a manager at Facebook, stated that the company’s tech roadmap extending to 2026 includes developing an AI system dedicated to recommendations that can manage the short video service Reels as well as traditional long-form videos.
Meta currently operates a different model for each category of its products, including features like Reels, Groups, and the main Facebook feed, as mentioned by Allison during the Morgan Stanley Technology Conference held in San Francisco.
Allison mentioned that the initial step in Meta’s tech roadmap involves updating their recommendation systems to operate using graphics processing units instead of traditional computer chips, aiming to enhance overall product performance efficiency.
With the increased interest in advanced language models over the past year, Meta’s leaders were surprised by the efficiency of these massive AI systems in analyzing vast amounts of data and managing various operations for comprehensive purposes like conversations, as noted by Allison.
Allison mentioned that Meta managed to innovate a massive recommendation model that can be applied to diverse products and established a new architecture for these model types last year, indicating that the company tested this system through the Reels platform.
The new architecture contributed to the development of the model, enabling Facebook to achieve up to a ten percent increase in watch time for Reels content via the Facebook app, showcasing the model’s ability to efficiently learn information from data compared to previous generations.
Meta is currently executing the third phase of restructuring its system architecture, which involves ensuring the efficiency of the technologies used and deploying them across a variety of products.
Allison said, “Instead of limiting ourselves to using Reels only, we are striving to work on an initiative to activate the entire video ecosystem through this unique approach, which will enable us to integrate a product that provides recommendations within the content delivered through this method.”
Meta has gathered massive sets of graphics processing units dedicated for use in expansive initiatives in generative artificial intelligence, such as those used in developing virtual assistants.
Meta’s projects in generative artificial intelligence include integrating advanced conversation technologies within its main platform, as well as experimenting with integrating an AI-powered conversation tool within groups.
