For the second time, Mark Zuckerberg revamps Meta’s AI program.
People familiar with the situation claim that amidst internal disagreements about the technology, Meta made an internal announcement regarding a reorganization of its artificial intelligence unit.
In recent months, Mark Zuckerberg, CEO of Meta, has been reorganizing the company’s artificial intelligence initiatives. He has now taken additional steps that risk exacerbating domestic unrest over the technology.
Two sources with knowledge of the situation stated that Meta informed its A.I. department, known as Meta Superintelligence Labs, internally on Tuesday that it would be divided into four teams. They stated that one group will concentrate on infrastructure, such as data centers and other A.I. hardware, one on products, one on A.I. research, and one on a possible potent A.I. known as “superintelligence.”
The reorganization is likely to be the last one for a while, according to the folks. According to the sources, the steps are meant to improve Meta’s organization so that it may advance toward its goal of superintelligence and create A.I. products more quickly in order to compete with other companies.
According to the sources, several A.I. officials are expected to depart. Since the A.I. team has expanded to thousands of employees in recent years, Meta is also considering reducing its size, perhaps by eliminating jobs or transferring staff to other areas of the organization, according to sources. They added that the downsizing is still up for debate and that no conclusions have been reached yet.
According to the source, the business is also actively considering employing third-party artificial intelligence models to power its A.I. products, which would represent a change from Meta’s current practice of solely relying on its own technology. Licensing “closed-source” models from other firms or building upon additional “open-source” A.I. models that are freely accessible are examples of such.
The shifts occur after months of upheaval and reorganization at Meta regarding its artificial intelligence approach. As the race to develop the most cutting-edge technology has come down to a handful of major players, Mr. Zuckerberg, 41, is prepared to make any investment and revolutionize his business to maintain his relevance in A.I. The public is paying close attention to how Meta will do in the face of the shifting winners and losers in the artificial intelligence competition.
After Meta had trouble promoting its newest product in June, Mr. Zuckerberg’s resolve was clear. models based on artificial intelligence. The business unveiled a superintelligence lab that month, with the goal of developing an AI that would surpass the human intellect capabilities. Alexandr Wang, the head of the startup Scale AI, was brought on as Meta’s new chief A.I. officer after the company invested $14.3 billion in it. Additionally, Meta sparked a poaching war in Silicon Valley by offering researchers from competing companies like OpenAI and Google pay packages worth nine figures.
Last month, during an investor call, Mr. Zuckerberg stated that he was banking on superintelligence to usher in “a new era of individual empowerment,” adding that A.I. has already enhanced Meta’s main advertising business. The business stated that its capital expenditures this year may reach $72 billion, with the majority of it going towards hiring AI scientists and constructing data centers.
The media was not given a statement by a Meta spokesperson. The Information previously covered some aspects of the restructuring.
Tensions have arisen since Mr. Zuckerberg established the superintelligence team under Mr. Wang. According to two persons familiar with the situation, Mr. Wang’s team is working on developing the company’s most potent A.I. model, which is referred to as a “frontier model.”
The next product from Meta has been discussed by the new squad. “Closed” A.I. model, which would mark a significant break from the company’s longstanding practice of “open sourcing” its models. An open-source A.I. model can be developed upon by other developers, but a closed model maintains the confidentiality of its underlying technology.
The new team has decided to build a new model from the ground up, abandoning Meta’s prior frontier model, dubbed Behemoth, according to the sources. One individual claimed that Behemoth’s release was postponed last spring following subpar performance tests.
Three persons familiar with the situation said that some members of the old guard have chafed at the new hires as Meta has spent billions to attract A.I. talent.
Meta appointed Shengjia Zhao, a co-creator of its ChatGPT chatbot and OpenAI researcher, to the position of chief AI scientist in July. According to one source, Mr. Zhao has been interrogating Meta’s former AI researchers and personnel about their previous work outside his office in recent weeks while interviewing them for potential new positions.
Two of the individuals said that Nat Friedman, the former head of GitHub, and Daniel Gross, who formerly ran a firm called Safe Superintelligence, will be in charge of creating new A.I. features under the products and applied research sector.
There has been staff turnover. Two individuals stated that Loredana Crisan, a vice president of generative A.I., is also leaving the firm to become Figma’s chief design officer at the software business. Earlier this year, Joelle Pineau, a prominent computer scientist at Meta, departed the business and joined Cohere, an AI startup. Angela Fan, a research scientist who worked on Meta’s open source A.I. model, dubbed Llama, recently left for OpenAI.
Some veteran AI leaders have remained at their positions. According to reports, Rob Fergus, who co-founded Meta’s AI research unit in 2014, will continue to serve as the day-to-day head of the company’s Fundamental AI Research lab, or FAIR. The FAIR division is in charge of promoting A.I. technology via open-source research. Mr. Wang is in charge of Ahmad Al-Dahle and Amir Frenkel, who have been working on generative A.I. solutions, and they are concentrating on strategic A.I. projects.
