Social media giant Meta is restructuring its AI department as it strives to achieve superintelligence.
Media outlets reported Monday that Alexandr Wang, founder of Scale AI and Meta's new chief AI officer, sent an internal memo outlining the restructuring of Meta Superintelligence Labs into four groups: TBD Lab, which will focus on foundation models such as Llama models; Fundamental AI Research or FAIR, which is an internal AI research lab; Products and Applied Research, a group that will incorporate models and research into consumer products; and MSL Infra, which will focus on infrastructure.
The restructuring is part of Meta's push toward building systems that exceed human intelligence. Meta has been on an aggressive hiring spree in the past few months, snapping up top AI researchers from companies including OpenAI, Anthropic and Google.
Benefits to restructuring
The restructuring is a good thing and has been well received, according to R "Ray" Wang, founder of Constellation Research.
From what I'm hearing internally, this is good.
R 'Ray' WangFounder, Constellation Research
"From what I'm hearing internally, this is good," Wang said. "Now they're going to streamline each group to figure out what they need because they were just doing random AI projects for a while."
The restructuring is also suitable for shareholders who are looking to see if Meta is focused, Wang continued.
"That's the important piece: Meta is extremely focused on AI, and they're not going to be one of the ones that falls behind," he said.
It's also clear that the restructuring enables the company to focus more on using AI for consumer and internal purposes.
"If you're a company like Facebook and you have all this money, it makes sense that you would develop AI for proprietary purposes for what you want to do," said Mark Beccue, an analyst at Omdia, a division of Informa TechTarget.
The company's pursuit was made clear during a second-quarter earnings call in July. Meta CEO Mark Zuckerberg told investors: "Meta's vision is to bring personal superintelligence to everyone -- so that people can direct it toward what they value in their own lives. We believe this has the potential to begin an exciting new era of individual empowerment."
Beccue added that while Meta has found success with open source in its Llama models, it is mainly not an enterprise player in the AI market. Therefore, the reorganization probably involves directing AI in a way that works best for Meta, he said.
"It sounds like refocusing the effort toward revenue generation," he said. "It could mean what happens to open source Llama, right? Does it get given to the Linux Foundation, or how long will they support that as it is right now?"
Some challenges
While Meta's restructuring makes sense given the social media giant's goals regarding closed models and superintelligence, it doesn't come without challenges.
"It's not going to be easy, even for Meta, even with all the talent, even with all the resources they have," said Chirag Shah, a professor in the Information School at the University of Washington in Seattle.
He said that vendors are running out of data to train the model when it comes to creating closed models. Others, such as Reddit, are guarding their data.
"If Meta or others want to keep building better models, they're going to have to find different ways to do that than just relying on more data," Shah said.
Moreover, GPT-5 is an example of how Meta's superintelligence goal is still far off. Shah said the model makes basic reasoning mistakes that a human wouldn't make.
"We still have a long way to go for superintelligence, because by definition, it's going to be better than the human brain," he said. "But we have difficulty even getting to that human level of intelligence. Not for everything, [but for] some of these basic challenges of reasoning, directing, explaining and taking actions."
Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence software and systems.
