After spending months hiring more than fifty AI researchers and engineers, Meta Platforms has halted hiring in its AI section, according to sources with knowledge of the situation.
Starting last week, the hiring freeze, which is part of a larger group reorganization, also restricts current staff from transferring between teams inside the division. The freeze’s duration was not discussed internally.
According to the persons, there may be an exemption to the ban on hiring outside parties, but they would require approval from Alexandr Wang, Meta’s chief AI officer.
The freeze was verified by a Meta representative, who described it as fundamental organizational planning: establishing a strong framework for the forthcoming superintelligence endeavors after enlisting personnel and carrying out annual planning and budgeting exercises.
Even while many of the leading AI companies have made large hiring efforts this year, Meta has frequently accelerated the recruitment fight by paying top researchers nine figures and employing so-called reverse acquisitions to deprive startups of important business executives. The size of investments made by major internet companies has alarmed analysts, who have cited Meta’s rapidly increasing stock-based compensation expenses as a possible danger to shareholder returns.
According to the people, Meta’s recent reorganization splits its AI efforts into four teams: one, called TBD Lab, is focused on superintelligence and houses many of the new hires; another is working on AI products; a third is working on infrastructure; and a fourth is devoted to projects with a longer time horizon and more exploration. The latter, known as Fundamental AI Research, is mainly unaffected by the restructuring.
The name Meta Superintelligence Labs, which encompasses the four divisions, reflects CEO Mark Zuckerberg’s current focus on creating computer systems that can execute cognitive tasks better than the most intelligent people.
AGI Foundations was the name of the team that worked on the most recent iterations of Meta’s large language models, or Llama. This spring, officials criticized the crew after the most recent Llama models failed to meet expectations. Its dissolution was a component of the reorganization.
According to internal messages seen by The Wall Street Journal, during the time of Meta’s most recent vesting date, August 15, at least three members of the former AGI Foundations team announced internally that they were departing the firm.
Zuckerberg took an active role in hiring AI experts following the April publication of the Llama model. Through extortion emails and WhatsApp chats, he reached out to staff members at OpenAI, Google DeepMind, and other laboratories, resulting in proposals that, in certain circumstances, totaled $100 million in compensation. His offer to one researcher, Andrew Tulloch, a co-founder of the Thinking Machines Lab, may have totaled up to $1.5 billion. (Tulloch turned down the offer.)
In order to get Wang, a co-founder of Scale AI, to lead Meta’s AI efforts, Zuckerberg paid $14 billion for a share in his business. Along with Daniel Gross, a co-founder of Safe Superintelligence, and Nat Friedman, the former CEO of GitHub, he courted them by giving Meta a share in their venture capital business.
For the endeavor, Meta had so far hired over 20 OpenAI researchers and engineers, at least 13 from Google, three from Apple, three from xAI, and two from Anthropic, for a total of more than fifty new hires as of mid-August.
Technology stocks have been selling down this week, partly due to growing investor concerns about the price of the tech giants’ AI rollout. In a research note dated August 18, Morgan Stanley analysts cautioned that the rapidly increasing stock-based remuneration that Google and Meta are offering to entice AI talent may jeopardize their capacity to return shares to shareholders. The analysts stated that excessive investment in talent might either boost AI innovations with enormous value creation or reduce shareholder value without producing any clear innovation improvements.
