Mark Zuckerberg Details Meta’s Plan for Self-Improving, Superintelligent AI

Meta CEO Mark Zuckerberg told investors that the newly formed Meta Superintelligence Labs is focused on building AI models that can self-improve—meaning they can learn from themselves without as much human input. The remarks came during a second-quarter earnings call on Wednesday.

“At some level, [it’s] not just going to be learning from people, because you want to build something that is fundamentally smarter than people,” Zuckerberg said. “So…you’re going to develop a way for it to improve itself. That is a very fundamental thing that’s going to have broad implications for how we build products and how we run the company.”

It’s one of a handful of comments the CEO has made during the past 12 hours that gesture at how Meta’s new research lab plans to compete with rivals like OpenAI, Google DeepMind, and Anthropic.

Early this morning, Zuckerberg posted a letter online and shared a video to Instagram Reels in which he said that Meta’s superintelligence efforts would center on building “personal superintelligence” that would give people tools for their own empowerment and to better the world. “This is distinct from others in the industry who believe superintelligence should be directed centrally towards automating all valuable work, and then humanity will live on a dole of its output,” he said.

Another key piece of Meta’s AI strategy is smartglasses, which Zuckerberg said he believes “are basically going to be the ideal form factor for AI.” Since they debuted in 2023, Meta has sold two million pairs of smart sunglasses made in partnership with Ray-Ban. Last September, the company showed off a futuristic prototype of augmented reality glasses, too. (The prototype, called Orion, isn’t expected to ship to consumers, but Meta has said it plans to build new devices over the next few years based on Orion research.)

For now, Meta appears to be making some distinction between AI that powers the monetization of its core products, like Instagram and WhatsApp, and superintelligent AI that could one day help power humanity’s future.

Zuckerberg’s remarks about superintelligence came on the heels of a better-than-expected earnings report. The company, which has lagged behind its competitors in the AI race, is spending billions of dollars to build out a small but mighty superintelligence lab that will focus on building frontier AI models.

Since publicly announcing the lab last month, Meta has been on an aggressive recruiting spree, bringing in key Silicon Valley operatives like Alexandr Wang, Nat Friedman and Daniel Gross, and offering AI researchers compensation packages as high as nine figures in order to lure them over to the lab. Wang was a part of an “acqu-hire” deal struck with Scale AI, an AI data-labeling startup that he cofounded. Former OpenAI researcher Shengjia Zhao has been named the head of Meta’s new research lab.

Meta revised its expectations for its capital expenditures for the year, increasing its forecast to $69 billion. Employee compensation is one of the biggest drivers of that, Meta’s chief financial officer Susan Li has said, as well as AI infrastructure. Despite the amount the company is spending on building out its AI products, it’s also forecasting a better-than-expected outlook for the third quarter, anticipating between $47.5 billion and $50.5 billion in revenue.