AI: Meta is reportedly building a new team to develop a superintelligence

An AI superintelligence is a kind of step towards Artificial General Intelligence (AGI). Meta is reportedly working on setting up a new superintelligence team. Mark Zuckerberg has personally taken over the search for talent. He is said to have invited several candidates to his home. A first team member already seems to have been found.

Meta is to take over Scale AI. The San Francisco-based company offers annotated data. Annotated means that it is provided with comments – which is necessary for AI training, for example. Originally, Scale AI mainly processed data from the field of autonomous driving. The company was founded by Alexandr Wang, among others. There are reports that Meta is paying 15 billion US dollars for Scale AI, Wang and the other employees, who will then automatically switch teams. Scale AI has 900 employees according to its own website.

Scale AI has a large number of contracts, many of which are with governments – also from Europe – as well as with the US military and the Department of Defense. Companies such as OpenAI, Microsoft and Cohere already rely on the company's services. In addition to providing data, Scale AI can also gain insights from data and offers its own AI models, for which Metas Llama has already been used. In the USA, some security agencies are experimenting with Defense Llama, a large language model based on Metas Llama, refined ("fine-tuned") by Scale AI.

However, Scale AI has also been criticized because the startup founded an independent agency, Scale Remotasks, where employees were hired to annotate the data – mostly in Southeast Asia and Africa. The working conditions are said to have been poor. Now anyone can apparently register and complete tasks online, for which they are then paid.

Zuckerberg wants to poach AI experts

As reported by the New York Times and Bloomberg, Meta has also made offers to dozens of other AI experts – including some who work at Google and OpenAI. Meta has the advantage of being able to cross-subsidize AI developments. Unlike OpenAI, for example, which only earns money with AI, Meta has the strong advertising business behind it. Meta has also had its own AI experts in-house for a long time, for example the AI science team Fair (Fundamental AI Research) is based in Paris, while the employees responsible for Meta AI and consumer applications are based in San Francisco.

According to Bloomberg, Zuckerberg himself is said to have been keen to put together the new team because he was dissatisfied with the reactions to Llama 4. In addition, the release of the largest model, Llama 4 Behemoth, has been postponed. The report also states that there is a WhatsApp group called "Recruiting Party" in which potential new employees are discussed.

According to their own statements, all major AI companies are working on developing an AGI. At Meta, however, chief AI scientist Yann LeCun doesn't think much of it. At the very least, he says that current AI models and generative AI in general are the wrong way to achieve such all-encompassing intelligence. LeCun talks about wanting to develop an AMI first, an Advanced Machine Intelligence, for which he and his team are developing Jepa, an AI model that learns from videos but is not generative.

LeCun is not alone in his concerns about the race for AI. Yoshua Bengio, also a Turing Award winner like LeCun, is concerned about security, for example, and has founded an AI security company. Anthropic has split off from OpenAI to develop a safe AI.

Joëlle Pineau, Meta's Vice-President for AI research, also left her job in April. She wrote in her farewell letter: "Today, as the world is fundamentally changing, the race for AI is accelerating and Meta is preparing for its next chapter, it is time to make room for others who want to continue this work." There might be some disagreement within Meta about where the AI journey should go.

(emw)