Meta Splits AI Lab, Shifts to Engineering Focus Amid Internal Chaos

Meta, the company behind Facebook, has recently announced a significant reorganization of its AI department, splitting its newly established Superintelligence Lab into four separate teams. This move marks the fourth time in six months that the company has adjusted its AI organizational structure, highlighting its ambitious and anxious stance in the AI race.

The reorganization, announced on August 20, involves redistributing many of the company's AI employees across the new teams. Notably, the team responsible for developing large models is discussing the possibility of transitioning to a closed-source model, which contradicts the open-source philosophy long advocated by Meta's Chief AI Scientist, a Turing Award winner.

Despite the aggressive recruitment of top AI talent from various companies, Meta is facing internal chaos and significant talent attrition. The frequent organizational restructuring has exacerbated tensions among employees, potentially leading to further instability within the company's technological operations.

The new structure includes four teams: TBD Lab, FAIR, PAR, and MSL Infra. TBD Lab, led by the former CEO of Scale AI, will focus on developing cutting-edge large models, including the next generation of the flagship Llama series. FAIR, which has been part of Meta for some time, will continue to focus on fundamental AI research, led by Robert Fergus, who recently returned from Google. PAR, co-led by the former CEOs of GitHub and Security Super AI, will aim to quickly convert AI technologies into user-friendly products. MSL Infra, headed by the Vice President of Engineering, will concentrate on building the infrastructure needed for AI training and inference, aligning with Meta's high annual capital expenditure plan.

This reorganization signals Meta's shift from a research-oriented strategy to an engineering-focused approach, aiming to accelerate the development and implementation of "superintelligence" through a full-stack layout. However, the marginalization of the Chief AI Scientist in this restructuring indicates a strategic shift away from open-source principles, which have been a cornerstone of Meta's AI strategy.

Meta's aggressive recruitment of AI talent has been met with internal chaos and organizational crises. Despite offering high salaries and benefits to attract top talent, the company is struggling with high turnover rates and a toxic work environment. The frequent restructuring and high-pressure performance metrics have created a culture of fear and mistrust, making it difficult for teams to collaborate effectively.

In response to the setbacks faced by its Llama 4 model, Meta's leadership has been attempting to reposition the company as a leader in "superintelligence" through organizational restructuring, significant investments, and the recruitment of top talent. However, the internal strategic fluctuations, organizational fractures, and talent attrition pose significant challenges to Meta's ability to compete in the AI race. While technology and talent can be acquired, building a strong company culture and organizational trust takes time and cannot be rushed. Meta's success in the AI competition will depend not only on technological breakthroughs but also on its ability to address the deep-seated organizational issues within the company.