Cracks are forming in Meta’s partnership with Scale AI

It has only been since June that Meta has invested $ 14.3 billion in the Data Label supplier scale AI, which resulted in CEO Alexandr Wang and various of the best executives of the Startup Meta Superintelligence Labs (MSL). But the relationship between the two companies already shows signs of fraying.

At least one of the managers who brought Wang to help run MSL – the former senior vice -president of the AI ​​of Genai Product and Operations scale, Ruben Mayer – left with the company after just two months, two people who are familiar with the case to WAN told.

Mayer spent about five years with aan scale over two stints. In his short time at Meta, according to those sources, Mayer supervised AI Data Operations Teams, but is not part of the TBD labs of the company -the core unit within Meta that is responsible for building AI Superintelligence, where Top AI researchers from OpenAI have landed.

However, Mayer disputes some details about his role and WAN told that his first position was “to set up the lab, with what was needed” instead of data, and that he “was part of TBD Labs from the first day” instead of the core of AI unit. Mayer also clarified that he ‘did not report directly [Wang]’And was’ very happy’ with his meta experience.

In addition to the personnel changes, Meta’s relationship with scale AI seems to be shifting. TBD Labs works together with external data labeling suppliers than scale AI to train its upcoming AI models, according to five people who are familiar with the issue. Those external suppliers are Mercor and Surge, two of the biggest competitors of scale AI, the people said.

While AI laboratories usually work with different suppliers of data suppliers – Meta has been working with Mercor and Surge since TBD Labs was spun – it is rare that an AI Laboratory invests so heavily in one data supplier. This makes this situation particularly remarkable: even with the investments of Meti Miljard Dollar from Meta, different sources said that researchers in TBD laboratories see the data from scale AI as low quality and have expressed a preference to work with golf and mercor.

Scale AI initially built its company on a crowdsourcing model that used a large, cheap workforce to process simple data tires, the process of tagging and annotating raw information to train AI models. But as AI models have become more advanced, they now require highly skilled domain experts, such as doctors, lawyers and scientists to generate and refine the data of high quality to improve their performance.

Although Scale AI has moved to attract these experts on the subject with his Bijterplatform, competitors such as Surge and Mercor have grown rapidly because their business models have been built on a high -paid talent from the start.

A meta spokesperson disputed the fact that there are quality problems with the product of Scale AI. Surge and Mercor refused to comment. Asked about the in -depth dependence on meta of competing data providers, a spokesperson for the AI ​​WAN scale focused on being first announcement From the investment of Meta in the startup, which quotes an extension of the commercial relationships of the companies.

The deals of Meta with data sellers from third parties probably means that the company does not place all its eggs in scale AI, even after having invested billions in the startup. However, the same cannot be said for scale AI. Not long after Meta announced his huge investment with scale AI, said OpenAi and Google that they would stop working with the data provider.

Shortly after losing those customers, dismissed scale AI 200 employees in data tennis Affairs in July, with the new CEO of the company, Jason Wore, partly blaming ‘shifts in market demand’. Wore said that scale AI would be staff in other parts of the company, including the sale of the government – the company has just one $ 99 million contract With the US Army.

Some initially speculated that the investment of meta in scale AI was real to lure Wang, a founder who had operated in the AI ​​room since Scale AI was founded in 2016 and who seems to help Meta to attract top AI talent.

Apart from Wang, there is an open question about how valuable scale is for Meta.

A current MSL employee says that several of the scale managers who have been transferred to Meta do not work on the Core TBD Labs team.

In the meantime, the AI ​​unit of Meta has become increasingly chaotic since he addresses Wang and a wave of top researchers, according to two former employees and a current MSL employee. New talent from OpenAi and Scale AI have pronounced frustration about navigating the bureaucracy of a large company, while the previous Genai team of Meta has seen its scope, they said.

Tensions indicate that the largest AI investment of Meta can so far be a rocky start, even though it had to tackle the AI ​​development challenges of the company. After the matte launch of Lama 4 in April, Meta -CEO Mark Zuckerberg was frustrated by the AI ​​team of the company, a current and a former employee told WAN.

In an attempt to turn things around and catch up OpenAi and Google, Zuckerberg hurried to close deals and launched an aggressive campaign to recruit top AI talent.

In addition to Wang, Zuckerberg has succeeded in getting the top AI researchers from OpenAi, Google DeepMind and Anthropic. Meta has also taken over AI Voice Startups, including Play AI and Waveforms AI, and has announced a partnership with the AI ​​Image Generation Startup, Midjourney.

To provide its AI ambitions with electricity, Meta recently announced various massive data center -Buildouts in the US, one of the largest is one $ 50 billion data center In Louisiana named Hyperion, named after a titan in Greek mythology that conceived the god of the sun.

Wang, who is not an AI researcher in the background, was seen as a somewhat unconventional choice to lead an AI lab. Zuckerberg has reportedly held conversations to bring in more traditional candidates to lead the effort, such as the Chief Research Officer of OpenAi, Mark ChenAnd tried to acquire the startups of Ilya Sutkever and Mira Murati. They all dropped.

Some of the new AI researchers who have recently brought in from OpenAI Al Meta left behindWired previously reported. In the meantime, many old members of the Genai unit of Meta have left in the light of the changes.

MSL AI researcher RISHABH AGARWAL is one of the newest, Post on X This week he would leave the company.

“The field of Mark and @alexandr_wang to build in the super intelligence team was incredibly attractive,” said Agarwal. “But in the end I choose to follow Mark’s own advice:” In a world that changes so quickly, the greatest risk that you can take is not a risk “.”

Then asked about his time at Meta and what his decision Dreef to leave, Agarwal refused to comment.

Director of Product Management for Generative AI, Chaya Nayakand research engineer, Rohan Varmahave also announced their departure from Meta in recent weeks. The question now is whether Meta can stabilize its AI activities and retain the talent it needs for his future success.

MSL has already started working on the next generation AI model. According to reports of Business insiderIt wants to end by the end of this year.

UPDATE: This story has been updated with comments from Mayer, who contacted WAN after publication.