eWEEK content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.
Mark Zuckerberg has officially revealed the team members of Meta’s new “superintelligence” division, following a months-long recruitment drive. Facebook’s parent company successfully poached seven researchers from OpenAI, two from Google DeepMind, one from Anthropic, and one from Sesame, an artificial intelligence hardware company.
Meta Superintelligence Labs: Who is – and isn’t – on the team, as of now
The team in Meta’s new AI research division, officially dubbed Meta Superintelligence Labs, will be led by Alexandr Wang, the 28-year-old co-founder of Scale AI, according to an internal memo seen by The Wall Street Journal. Meta invested $14.3 billion for a 49% stake in Wang’s data-labelling startup earlier this month and, as part of the deal, he joined Meta. “I consider him to be the most impressive founder of his generation,” Zuckerberg said in the memo.
While Wang is the Chief AI Officer, he leads alongside former GitHub CEO Nat Friedman, who will work mainly on AI products and applied research. Another 11 new hires were listed in the memo, while Zuckerberg said more “great people at all levels (are) joining this effort in the coming weeks.”
Conspicuously absent from the memo was Daniel Gross, Friedman’s business partner and CEO of Safe Superintelligence, the AI startup co-founded by OpenAI’s Ilya Sutskever. Rumours have circulated that Zuckerberg had hoped to recruit Gross after Sutskever declined both acquisition and hiring overtures.
There was also no mention of Lucas Beyer, Alexander Kolesnikov, and Xiaohua Zhai, who all left OpenAI Zurich for Meta together last week. Nevertheless, Sam Altman did say on his brother Jack Altman’s podcast that “none of (OpenAI’s) best people” took an offer from Meta.
All teams working on AI will now fall under Meta Superintelligence Labs, which includes the FAIR fundamental AI research team, the team developing the Llama models, and the team that develops the Meta AI products, such as its AI assistant. There is also a new lab “focused on developing the next generation of our models,” Zuckerberg said.
Meta Superintelligence Labs will be working on Llama 4.1 and 4.2, as confirmed in the memo, despite rumours that Meta is considering de-investing in Llama after a lacklustre reception to its roll-out. At the same time, the researchers will be developing new models “to get to the frontier in the next year or so.”
Why Meta is ‘uniquely positioned’ to deliver on superintelligence
Superintelligent AI systems are those capable of outperforming humans on nearly all tasks. Zuckerberg said in the memo that Meta is “uniquely positioned” to deliver it, as it has enough cash to acquire the significant compute resources required, a track record of delivering products to billions of people, experience in AI hardware such as smart glasses, and a company structure that allows it to take big swings.
“I’m optimistic that this new influx of talent and parallel approach to model development will set us up to deliver on the promise of personal superintelligence for everyone,” he wrote in his memo.
Will Zuckerberg’s furious recruitment drive change its AI fortune?
Meta Superintelligence Labs is the result of a furious recruitment drive from Meta that has been going on for “the past few months,” according to the WSJ’s memo. The CEO has been getting his hands dirty, scouring AI research papers and compiling a list of the most-cited authors as potential candidates, whom he then interviews himself and offers signing bonuses of up to $100 million.
While his ultimate goal is for Meta’s AI models and products to become ubiquitous, Zuckerberg has recently been working hard to keep pace with key rivals OpenAI and Google. It was on the back foot in the early years due to the company’s initial focus on supervised learning, where large labelled datasets drive performance, over reinforcement learning, where agents learn by interacting with environments through trial and error.
Another error was its decision to open-source its models, with the intention of making its technology the go-to choice for developers. This backfired when DeepSeek released models based on Llama that were more advanced and less expensive.
Fast forward to April, and the release of Meta’s latest large language model, Llama 4, was delayed at least twice due to its underperformance in technical benchmarks and conversationality, according to The Information. When it was released to little fanfare, experts pointed out that the version listed on the LMArena leaderboard differed from what was publicly released, suggesting Meta had submitted a tweaked model optimised for benchmarks to improve its image.
Meta has been recruiting for a while to try to change its fortune, as well as replace the growing number of researchers who are leaving for competitors, but its less-than-stellar AI reputation has not made it easy. Meta’s AI chatbots have been engaging in seriously inappropriate conversations, and the company has been entangled in legal battles with creatives accusing it of stealing their work for training data.
