Facebook is working on a single AI model to power all video recommendations

of Meta major investment in artificial intelligence includes developing an AI system designed to power Facebook’s entire video recommendation engine across all of its platforms, a company executive said Wednesday.

Tom Allisonthe Facebook executive, said part of Meta’s “technology roadmap” going to 2026 includes developing an artificial intelligence recommendation model that could power both the TikTok-like short video service Reels and more traditional, longer videos.

To date, Meta typically uses a separate model for each of its products, such as Reels, Groups and the main Facebook Feed, Allison said on stage at the Morgan Stanley Technology Conference in San Francisco.

As part of Meta’s ambitious foray into AI, the company is spending billions of dollars on Nvidia graphics processors, or GPUs. They have become the primary chips used by AI researchers to train the kinds of large language models used to power OpenAI’s popular ChatGPT chatbot and other generative AI models.

Allison said “phase 1” of Meta’s technology roadmap involves switching the company’s current recommendation systems to GPUs from more traditional computer chips, helping to improve overall product performance.

As interest in the LLM grew last year, Meta executives were struck by how these large AI models could “process a lot of data and all kinds of very general-purpose activities like chat,” Allison said. Meta came to see the opportunity for a giant recommendation model that could be used across products, and by last year had built “this kind of new model architecture,” Allison said, adding that the company had been testing it on Reels.

This new “model architecture” helped Facebook get an “8% to 10% gain in Reels watch time” in Facebook’s core app, which Allison said helped prove that the model “learns from the data much more effectively from the previous generation. “

“We’ve really focused on investing more to make sure we can scale these models with the right kind of hardware,” he said.

Meta is now in “Phase 3” of its system re-architecture, which includes attempts to validate the technology and push it into multiple products.

“Instead of just powering Reels, we’re working on a project to power our entire video ecosystem with this single model, and then can we add our recommended Feed product to be served by that model as well,” Allison said. “If we get this right, not only will the recommendations be somehow more engaging and more relevant, but we think their responsiveness can be improved as well.”

Illustrating how it would work if successful, Alison said: “If you see something you’re interested in on Reels and then go back to the Feed, we can show you more similar content.”

Allison said that Meta has built up a huge stockpile of GPUs that will be used to support broader generative AI efforts, such as the development of digital assistants.

Some generative AI projects that Meta is considering include incorporating more sophisticated chat tools into its main channel, so that a person who sees a “recommended post about Taylor Swift” might “easily just click a button and say: “Hey Meta AI, tell me more about what I’m seeing with Taylor Swift right now.”

Meta is also experimenting with integrating its AI chat tool within Groups, so a member of a Facebook baking group could potentially ask a question about desserts and get an answer from a digital assistant.

“I think we have the opportunity to put generative AI into a kind of multiplayer user experience,” Allison said.

I’M WATCHING: The full CNBC interview with Meta’s Nick Clegg

Don’t miss these stories from CNBC PRO: