Cracks are forming in Meta’s partnership with Scale AI

Meta has invested only $ 14.3 billion in the AI ​​Data Signs Summer scale, where the CEO of Alexandr Wang and many senior executives at the start starting to run Meta Superinteigence Labs (MSL). But the relationship between the two companies already shows signs of fading.

Robin Mayer, Robin Mayer – said at least one of the executives to help run the MSL – Scale Scale Enclude Product Product and Operations, Ruben Mayer – just two months after the company.

Mayer spent about five years with the scope of artificial intelligence across two tasks. In a short time in Meta, according to those sources, Mayer supervised AI’s data operations teams, but he was not part of the company’s TBD Labs – the basic unit within the Meta charged with the construction of SuperIntelligence, where the best researchers from AI fell from Openai.

However, Mayer opposes some details about his role, and told Techcrunch that his initial position is “helping to prepare the laboratory, with everything required” instead of data, and that he is “part of the TBD laboratories from the first day” instead of excluding it from the basic AI unit. Mayer also explained that he “did not submit a direct report to [Wang]He was very happy with his descriptive experience.

Besides employee changes, it appears that the Meta relationship with AI is the scale is transformed. TBD Labs works with third -party data classification sellers other than Scale AI to train the upcoming artificial intelligence models, according to five people who are familiar with this issue. The people said that these third -party sellers include Mercor and TruPGe, one of the largest AI’s competitors.

While Ai Labs usually works with many signs of data signs – Meta has been working with Mercor and Trashlge since TBD LABS – it is rare to invest the artificial intelligence laboratory significantly in one data seller. This makes this situation in particular noticeable: even with the investment that extends millions of dollars in Meta, many sources said that researchers in TBD laboratories see Scale AI as low quality and have expressed preference for work with translation and MERCOR.

Scale AI initially built its business on the collective outsourcing model that used a large -cost large -cost workforce to deal with signs of simple data, which is the process of laying raw information and explaining it to train artificial intelligence models. But since artificial intelligence models have grown more sophisticated, they now require high skills experts-such as doctors, lawyers and scientists-to create and refine the high-quality data needed to improve their performance.

Although Scale AI has moved to attracting these topics experts through its external platform, competitors such as GRGE and Mercor are growing quickly because their business models were built on the basis of high -wage talent from the beginning.

A Meta spokesman is the fact that there are quality problems with the AI ​​Scale product. The boom and the mirkor refused to comment. When asked about relying on Meta deepening on competing data providers, AI spokesman for Techcrunch to its initial announcement of Meta’s investments in startup, which indicates an expansion in the commercial relationship of companies.

Meta deals with third -party data sellers may mean that the company does not lay all its eggs in the AI ​​range, even after investing billions of dollars in starting. However, it is not possible to say the same with regard to the artificial intelligence scale. It was not long ago that Meta announced its massive investment with Scale AI, Openai and Google said they would stop working with the data provider.

Shortly after these customers lost, Scale Ai has handed over 200 employees in their data classification in July, with the new CEO of the company, Jason Droege, partially blames changes for “market demands”. Drewig said that Scale AI will work in other parts of business, including government sales – the company has obtained a $ 99 million contract with the US military.

Some preferred at the beginning that Meta’s investment in Scale AI was really the temptation of Wang, a work founder in artificial intelligence space since Scale Ai was founded in 2016 and it appears to be helping Meta to attract the best talents of artificial intelligence.

Regardless of Wang, there is an open question about the value of the scale.

One of the current MSL employees says that many CEOs who were brought to Meta are not working on the main TBD Labs team.

Meanwhile, the unit of Amnesty International in Meta has become increasingly chaotic since Wang brought and a wave of senior researchers, according to former employees and current employee in MSL. They said that the new talents from Openai and Scale Ai expressed their frustration with the movement of the bureaucracy of a large company, while the former Genai team in Meta witnessed its limited scope.

Tensions indicate that the largest investment of artificial intelligence in Meta so far may be a rock start, although it was supposed to address the challenges of developing the company’s artificial intelligence. After the fading Llama 4 launch in April, the CEO of Meta Mark Zuckerberg became disappointed by the company’s AI team, said one of the current employees and one former employee of Techcrunch.

In an attempt to convert matters and join Openai and Google, Zuckerberg rushed to conclude deals and launched an aggressive campaign to employ the best talents of artificial intelligence.

After Wang, Zuckerberg managed to obtain the major artificial intelligence researchers from Openai, Google Deepmind and Anthropic. Meta also acquired Amnesty International Voice Company including Play Ai and WaveForms AI, and announced a partnership with AI Image GENERATIAT, Midjourney.

To operate the aspirations of artificial intelligence, Meta recently announced many constructions of the huge data center throughout the United States, a $ 50 billion data center in Louisiana called Hyperion, named after Titan in Greek myths that generate the deity of Sun.

Wang, who was not an artificial intelligence researcher in the background, was seen as a somewhat unconventional option to lead the artificial intelligence laboratory. According to Zuckerberg, there were talks to bring more traditional candidates to lead this effort, such as the chief research official at Openai, Mark Chen, and tried to get the startups in Elia Soskfar and Mira Murrati. All of them refused.

Some new researchers who recently brought from Openai Meta have already left, previously reported. Meanwhile, many members of the Genai unit left in Meta in light of the changes.

MSL AI Rishaabh Agarwal is among the latest, as he posted on X this week that he would leave the company.

“The stadium was from Mark and @alexandr_wang to build on Superinteldignment team was incredibly convincing.” “But in the end, I choose Mark’s advice:” In a world that changes very quickly, the biggest danger you can carry is not risk. “

He then asked about his time in Meta and what prompted his decision to leave, and he refused to comment.

The Director of Products Management in the Instelligence Intelligence, Chaya Nayak, and Research Engineer, Rohan Pharma, announced their departure from dead in recent weeks. The question now is whether Meta can stabilize artificial intelligence and keep the talent it needs for its success in the future.

The MSL has already started working on the next generation AI. According to reports from Business Insider, it is aimed at launching it by the end of this year.

Update: This story was updated with comments from Mayer, which communicated with Techcrunch after publication.