Meta Disbands Llama 4 Development Team After Lackluster Performance
Key Llama Researchers Resign Amid "Morale Undermined by Restructuring"
Push Accelerates as Meta Recruits 50 for Superintelligence Lab
Meta Platforms, the parent company of Facebook (hereafter referred to as Meta), reportedly halted external talent recruitment for its "Meta Superintelligence Research Lab" starting last week. According to Yonhap News, Meta confirmed the recruitment freeze to The Wall Street Journal (WSJ) on the 20th (local time), stating, "We are building a solid structure after securing personnel and establishing annual budgets and plans for the new superintelligence project." However, it was reported that exceptions could be made if approved by Alexander Wang, the Chief AI Officer leading the lab.
Meta has halted external talent recruitment for the "Meta Superintelligence Research Lab." Reuters Yonhap News
Mark Zuckerberg, CEO of Meta, officially announced the establishment of the "Meta Superintelligence Research Lab" last month, with the goal of developing AI that surpasses human intelligence. The lab consists of four teams: the "TBD Lab," which oversees large language models (LLMs); "FAIR," an internal organization that has been researching AI for over a decade; a product and applied research team; and an infrastructure team. Most of the external hires so far have been assigned to the TBD Lab.
Meta has aggressively pursued talent for its superintelligence lab, offering astronomical compensation packages. In June, the company invested $14.3 billion (about 19.6 trillion KRW) in the startup Scale AI and brought in its founder and CEO, Alexander Wang. Meta also recruited Daniel Gross, co-founder of Safe Superintelligence (SSI) established by OpenAI co-founder Ilya Sutskever, as well as Nat Friedman, former CEO of the developer platform GitHub. Recently, Meta reportedly secured more than 50 external AI researchers and developers, including over 20 from OpenAI, at least 13 from Google, three from Apple, three from xAI, and two from Anthropic. Some of these hires were reportedly offered compensation packages worth up to 100 billion KRW.
In April, Meta released its LLM model Llama 4, but it was deemed below expectations by company executives. At the time, Zuckerberg directly expressed his dissatisfaction to the development team, calling the situation "unacceptable." As a result, the AGI Foundation Team, which had led LLM development, was disbanded during the recent restructuring. In addition, it was reported that some employees, feeling uneasy about continuous organizational changes, left Meta. Of the 14 co-authors listed on the Llama model research paper published in 2023, 11 have resigned.
Tijmen Blankevoort, a former Llama model researcher, wrote on the internal message board before leaving the company, "I haven't met anyone in the AI department who is truly enjoying themselves," adding, "Fear within the organization is spreading like metastatic cancer." He further criticized, "Frequent performance evaluations and repeated restructuring are undermining morale and creativity," and "Most people don't even know what our mission is."
Zuckerberg has stated, "We will take a different approach from other companies in the industry when developing superintelligent AI," and added, "Rather than focusing AI on labor automation, we will develop it in ways that help people's personal lives."
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
