An executive from Facebook has stated that the company is developing a massive AI model to drive all video recommendations.

• On Wednesday, Facebook's head, Tom Alison, announced that the company is developing an AI model to drive its entire video system.
• At the tech conference in San Francisco, Alison revealed that the model is included in Meta's "technology roadmap for the years 2026."
• Recently, Mark Zuckerberg, CEO of Meta, revealed that the company is investing billions of dollars in Nvidia's graphics processing units to enhance its AI capabilities.

An AI system is being developed by Facebook to power its entire video recommendation engine across all platforms, a company executive announced on Wednesday.

Facebook's CEO, Tom Alison, revealed that Meta's technology roadmap for the next six years includes the development of an AI recommendation model that will be used for both Reels, the company's TikTok-like short video service, and longer videos.

Alison stated at Morgan Stanley's tech conference in San Francisco that Meta has traditionally employed a separate model for each of its products, including Reels, Groups, and the core Facebook Feed.

Nvidia GPUs have become the primary chips used by AI researchers for training large language models, which power popular chatbots like ChatGPT and other generative AI models. Meta, the parent company of Facebook, has been spending billions of dollars on these GPUs as part of its ambitious AI initiatives.

Alison stated that Meta's tech roadmap's first phase involved replacing the company's existing recommendation systems with GPUs instead of traditional computer chips, resulting in enhanced product performance.

Last year, the popularity of LLMs increased significantly, prompting Meta executives to realize the potential of these large AI models to handle vast amounts of data and perform a wide range of general-purpose tasks, such as chatting. As a result, Meta considered building a massive recommendation model that could be used across various products. By the end of the year, the company had successfully developed a new model architecture and tested it on Reels.

The new "model architecture" helped Facebook achieve a 8% to 10% increase in Reels watch time on the core Facebook app, proving that it was more efficient in learning from data than the previous generation, as stated by Alison.

He stated that they have concentrated on investing in hardware to ensure the scalability of their models.

Meta is currently in the "phase 3" of its system re-architecture, which entails verifying the technology and integrating it into various products.

"Alison stated that instead of just powering Reels, the team is working on a project to power the entire video ecosystem with a single model. Additionally, they plan to add their Feed recommendation product to be served by this model. If they succeed, they believe the recommendations will be more engaging and relevant, and the responsiveness of them will improve."

Alison stated that by showcasing similar content on the Feed after discovering something interesting on Reels, we can provide a more personalized experience.

Alison stated that Meta has amassed a significant collection of GPUs, which will aid its broader AI initiatives, including the creation of digital assistants.

Meta is considering integrating advanced chat tools into its core Feed, allowing users to easily request more information about recommended posts, such as "Taylor Swift."

Facebook is testing the integration of its AI chatbot within groups, allowing members of a baking group to ask about desserts and receive an answer from a digital assistant.

Alison stated that she believes there is an opportunity to incorporate generative AI into a multiplayer consumer environment.

WATCH: CNBC's full interview with Meta's Nick Clegg