Meta builds applied AI organisation to improve model performance

Meta is building a new applied AI engineering organisation as part of its drive toward superintelligence. Reportedly, the new group will be led by Vice President of Reality Labs Maher Saba and will directly report to Chief Technology Officer Andrew Bosworth.

The group will be closely working with the Meta Superintelligence Labs, which is building its innovative AI models under former Scale AI chief Alexandr Wang.

Meta’s AI engineering organisation

The new organisation will feature unusually flat management, with manager-to-employee ratios of up to 1:50. This reflects CEO Mark Zuckerberg’s broader strategy of “elevating individual contributors and flattening teams,” which he highlighted in Meta’s latest earnings call.

According to Wall Street Journal, Saba told employees that building great AI models requires more than researchers and compute power, it also depends on strong internal tooling and robust data pipelines.

There will be two main teams, one of which will be in charge of developing interfaces and internal tools, while the other team will be focused on feeding AI models with high-quality data.

According to an internal memo, Saba emphasised that the goal is "to take existing AI models, which are already quite good, and turn them into market-leading AI models through the application of advances in reinforcement learning and post-training techniques."

Speeding up AI development, Meta's new AI engineering organisation can improve performance in a more efficient way. This new structure is similar to those used at other Big Tech companies, like Nvidia, which also utilises flat structures to enable individual contributors.