Meta has reportedly reorganized its recently formed ‘superininformigence’ lab into four units focapplyd on research, products, infrastructure, and advanced model development, while considering possible staff reductions, according to Bloomberg and The New York Times.
In an internal memo, Meta’s new chief AI officer, Alexandr Wang, detailed a major reorganization within the company’s AI division. The lab will now be split into four groups: TBD Lab, focapplyd on large language models like Llama; Fundamental AI Research, or FAIR, the company’s long-running research arm; a Products and Applied Research team; and MSL Infra, which oversees the massive infrastructure behind Meta’s AI push.
Wang will take charge of TBD Lab himself. Robert Fergus will remain at the helm of FAIR, Aparna Ramani will lead MSL Infra, and former GitHub CEO Nat Friedman will run the applied research unit.
Generative AI unit scrapped
The AGI Foundations team, which had been focapplyd on generative AI, has been dismantled. Its leaders, Ahmad Al-Dahle and Amir Frenkel, will shift to strategic projects within the new superininformigence unit and report directly to Wang. Connor Hayes, who previously led AI product development, was earlier relocated to the Threads team.
Internal friction over pay
Meta’s AI division has grown to several thousand employees, and sources state the company is considering scaling back, which could involve cutting jobs or shifting staff to other areas. Some senior executives are also expected to depart.
Within the company, longtime employees have become frustrated becaapply newcomers are being offered pay packages in the nine figures. This has created resentment and deepened divisions across teams.
Pivot on AI models
Meta has also abandoned plans for its high-profile model, Llama 4 Behemoth. Instead, Wang’s team is starting from scratch to build a new system that favors a closed-source design. This marks a major shift from the company’s multi-year commitment to open source.
The company is also considering utilizing third-party AI models, signaling a significant relocate away from its traditional strategy of relying solely on in-hoapply development.
Article edited by Jerry Chen
Leave a Reply