According to Zhitong Finance APP, due to disappointment with Meta (META.US) in the field of artificial intelligence, CEO Mark Zuckerberg is personally assembling an elite team to fully accelerate the research and development of Artificial General Intelligence (AGI). According to insiders, this secret new team is internally referred to as the "superintelligence group." Zuckerberg has set a goal: to make Meta a leader in the AGI field. He believes that compared to other tech giants, Meta not only has the capability but also the responsibility to be the first to achieve AGI—machines that can match human abilities across various tasks. Once AGI is achieved, its capabilities will be integrated into all of Meta's products, including social media, communication platforms, AI chat assistants, and AI tools like smart glasses. Zuckerberg is personally leading the team recruitment, aiming to recruit about 50 experts, including a new AI research director. He will personally conduct initial contacts and follow up throughout the process. To facilitate close collaboration, he has adjusted the seating layout at Meta's Menlo Park headquarters to have new members sit near him. Over the past month, he has frequently held meetings at his homes in Lake Tahoe and Palo Alto, personally recruiting AI researchers, infrastructure engineers, and entrepreneurs. While building the core team, Meta plans to make a multi-billion dollar investment in the data service platform Scale AI, which will be the largest external investment in Meta's history. The deal will value Scale AI at $28 billion. According to the Bloomberg Billionaires Index, this transaction will increase the stock value of Scale's founder Alexandr Wang to over $5 billion, and co-founder Lucy Guo will also become a billionaire. After the deal is completed, Wang is expected to join Meta's superintelligence group. Zuckerberg has publicly stated that AI is Meta's priority strategy. According to those close to him, he has returned to "founder mode" in the past two months, becoming increasingly hands-on in management. This shift stems from his strong dissatisfaction with the quality and response of Meta's latest large language model, Llama 4 (used to power chatbots, etc.). The latest version released in April disappointed Zuckerberg, and he has repeatedly told Meta insiders that he hopes to provide the best AI products in terms of total usage and performance by the end of the year. Insiders have revealed that his demands have put pressure on employees focused on AI, forcing them to work nights and weekends to meet these goals. However, these individuals have stated that the performance of these models has been questioned both internally (by Meta's own leadership) and externally, with developers believing they have over-promised and under-delivered. Meta has even postponed the release plan for its largest model "Behemoth," which claims to surpass competitors like OpenAI, Anthropic, and Google, due to leadership concerns about its limited improvements. These setbacks have prompted Zuckerberg to become deeply involved and determined to establish a new team. He has created a WhatsApp group among executives called "Recruiting Party" to discuss talent recruitment around the clock He firmly believes that Meta's strong advertising business cash flow (unlike competitors that rely on financing) is sufficient to support its massive investments in AI (with capital expenditures planned in the hundreds of billions this year, and expected to reach trillions in the future), and told potential candidates that Meta has the capability to build data centers with world-class computing power. However, Meta's investment in Scale AI may trigger antitrust scrutiny from the Federal Trade Commission (FTC). The agency is currently filing a lawsuit regarding Meta's acquisition of Instagram and WhatsApp. Other tech giants (Google, Microsoft, NVIDIA) are also facing scrutiny over AI-related transactions. Experts are concerned that if Meta invests in Scale AI, it may limit competitors' access to key services on the platform, especially considering that Meta hopes its LLM will become the industry standard, while Scale currently also serves Meta's competitors. Llama Predicament Meta's AI ambitions hinge on its Llama models, which support its AI chatbots and help train algorithms for delivering targeted content and ads to social media users. Meta has also open-sourced Llama, meaning the blueprint for the model is public, allowing others to build upon it. The company's hope is that Llama will become the foundation for global AI products, much like Google's Android provided a generation of software products for mobile devices. However, the recent iterations of Llama have not received a positive response. Even Meta employees feel that some of the successes publicly claimed by Zuckerberg are overly optimistic. Many users have stumbled upon Meta's AI products (including its chatbots) through the company's social media apps rather than actively seeking them out, and Llama's performance benchmarks have faced criticism from external reviewers. The release of Llama 4 comes after Meta has experienced years of leadership changes and shifts in priorities in AI. After OpenAI's ChatGPT set a new benchmark for AI in 2022, Meta quickly redirected talent and resources to establish a new, more consumer-focused AI team, GenAI. This team's mission is to translate years of AI-related research by Meta scientists into actual consumer products and the Llama model. But this adjustment has come with growing pains. Some insiders believe that Meta's initial consumer-focused AI efforts—including creating AI characters based on celebrities like Snoop Dogg and Tom Brady—seem frivolous compared to the higher goals (such as AGI) pursued by competitors like OpenAI. Concerns that Meta did not sufficiently recognize the disruptive potential of AGI have led to the departure of several executives. This turnover continues as Meta undertakes a comprehensive reform of its FAIR and GenAI departments this year As Meta's Chinese counterparts develop open-source artificial intelligence tools that some startups choose to collaborate with, the pressure will only increase. A Chinese startup named DeepSeek released a state-of-the-art open-source AI model in January, reportedly at a development cost that is only a fraction of that of American models. In March, South Korean AI chip startup FuriosaAI rejected Meta's proposal to acquire the company for $800 million. According to people familiar with the deal negotiations, Meta was interested in FuriosaAI's experts and engineering talent, hoping to bring them in to improve its own chip development efforts. This failed acquisition has raised questions about Meta's own efforts to manufacture custom high-performance chips for AI workloads and its talent pipeline focused on infrastructure