Recently, Yang Likun, who served as Meta's Chief AI Scientist, disclosed that last year, Meta undertook substantial adjustments to its AI strategy. A severe divergence of opinions emerged between the management and the technical team regarding research and development directions. As a result, the generative AI department found itself gradually sidelined, which in turn led to a significant exodus of core talent.

Meta's founder, Zuckerberg, insisted that the department expedite its R&D progress. This insistence, however, led to a deadlock in communication between the management and the technical team. The technical team felt that management was turning a blind eye to innovative ideas, thereby stifling the momentum of innovation.

In the market, Meta's Llama series models underperformed. The Llama 4, released in April 2025, was a notable flop and was even accused of manipulating test benchmarks. Yang Likun admitted that the team had indeed tampered with data results, a move that enraged Zuckerberg. Subsequently, the department was completely marginalized, and a large number of employees departed.

This controversy has laid bare Meta's internal conflicts and the technical challenges it faces in implementing its AI strategy.