Meta's AI ambitions have hit significant roadblocks in recent months, threatening the company's position in the increasingly competitive artificial intelligence landscape.
The troubles began in April when Meta faced accusations of manipulating AI benchmarks to make its recently released Llama 4 models appear more capable than they actually were. Critics pointed out that Meta had submitted a specially optimized version of its Llama 4 Maverick model to the LM Arena benchmark, different from the publicly available version. While Meta executives denied training on test sets, the controversy damaged the company's technical credibility.
Compounding these issues, Meta has now delayed the release of its flagship 'Behemoth' AI model multiple times. Originally scheduled for an April unveiling at LlamaCon, then pushed to June, the 2-trillion-parameter model has now been postponed until fall 2025 or later. According to reports, Meta engineers are struggling to achieve significant improvements over previous models, raising internal concerns about whether Behemoth's performance will justify a public release despite the company's substantial investments.
In response to these setbacks, CEO Mark Zuckerberg is taking direct control of Meta's AI strategy by personally assembling a new 'superintelligence' team. This elite group of approximately 50 AI researchers and engineers will be tasked with developing artificial general intelligence (AGI) capabilities that could eventually surpass human intelligence. Zuckerberg has reportedly rearranged Meta's headquarters to position this team near his office and is offering unprecedented compensation packages—some reportedly reaching nine figures—to attract top talent.
The restructuring effort includes a reported $14.3 billion investment in Scale AI, bringing its founder Alexandr Wang into Meta's AI leadership. This aggressive talent acquisition strategy comes after Meta's existing AI research team has seen significant departures, including 11 of the 14 authors of the original Llama paper.
Meta's struggles and subsequent reorganization highlight the intense pressure facing major tech companies in the AI arms race. With competitors like OpenAI, Google, and Anthropic making rapid advances, Zuckerberg's hands-on approach signals how critical AI development has become to Meta's long-term strategy and competitive position in the technology landscape.