Meta’s Investment in Scale AI: A Rocky Relationship Revealed
In a surprising move back in June, Meta made a significant investment of $14.3 billion in the data-labeling vendor Scale AI. This investment led to the appointment of CEO Alexandr Wang and several top executives from Scale AI to head Meta Superintelligence Labs (MSL). However, the once-promising relationship between the two companies is beginning to show signs of strain.
One of the executives, Ruben Mayer, who was brought over from Scale AI to assist in running MSL, has already left Meta after just two months with the company. Sources familiar with the matter revealed that Mayer’s role at Meta involved overseeing AI data operations teams but did not include participation in TBD Labs, Meta’s core unit responsible for building AI superintelligence.
Despite conflicting reports about Mayer’s role, it is evident that Meta’s reliance on Scale AI is shifting. Reports suggest that TBD Labs is now working with third-party data labeling vendors, such as Mercor and Surge, in addition to Scale AI. While it is common for AI labs to collaborate with multiple vendors, the preference for Surge and Mercor over Scale AI among TBD Labs researchers is notable, given Meta’s substantial investment in the latter.
Scale AI, initially known for its crowdsourcing model for data labeling, has faced challenges as AI models demand higher-quality data from domain experts. Competitors like Surge and Mercor, which prioritize skilled talent, have gained traction in the market. Despite claims from a Meta spokesperson that there are no quality issues with Scale AI’s product, the company’s deepening reliance on competing vendors suggests otherwise.
Following Meta’s investment, Scale AI experienced setbacks, including layoffs and losing customers like OpenAI and Google. These developments have raised questions about Scale AI’s value to Meta and the overall success of the investment. Issues within Meta’s AI unit, including bureaucratic challenges and scope limitations, further complicate the situation.
In response to these challenges, Meta has made aggressive moves to recruit AI talent and expand its AI capabilities through acquisitions and partnerships with AI startups. The company’s ambitious AI plans are supported by massive data center buildouts, such as the $50 billion data center project named Hyperion in Louisiana.
The rocky start to Meta’s largest AI investment highlights the complexities and challenges of developing cutting-edge AI technologies. As Meta strives to compete with industry leaders like OpenAI and Google, the company faces internal tensions and external uncertainties that may impact its AI development trajectory.
New Wave of AI Researchers Depart from Meta
Recent reports from Wired have revealed that a number of newly recruited AI researchers from OpenAI have decided to leave Meta, adding to the exodus of talent from the company. Additionally, several longstanding members of Meta’s GenAI unit have also chosen to depart amidst organizational changes.
One of the departing researchers is Rishabh Agarwal from Meta’s Superintelligence team at MSL. Agarwal took to X this week to announce his decision to leave the company, citing the changing landscape and the importance of taking risks.
In his statement, Agarwal mentioned the compelling pitch from Mark and @alexandr_wang to work on the Superintelligence team but ultimately decided to heed Mark’s advice on taking risks. When asked about his time at Meta and the reason for his departure, Agarwal chose not to comment.
Joining Agarwal in leaving Meta are Chaya Nayak, Director of product management for generative AI, and Rohan Varma, a research engineer. Their departures raise questions about Meta’s ability to stabilize its AI operations and retain top talent for future endeavors.
Despite the talent drain, MSL is forging ahead with its next-generation AI model, as reported by Business Insider. The lab aims to launch this new model by the end of the year, signaling a commitment to innovation despite the recent departures.
Update: This article has been revised to include comments from Mayer, who contacted JS following the initial publication.