Meta has announced its largest reorganization of artificial intelligence (AI) operations, aiming to accelerate progress toward "superintelligence," where AI surpasses human intellectual abilities. This came through an internal memo from Alexandr Wang, 28, head of Meta Superintelligence Labs (MSL), who outlined a restructuring into four key teams: research, training, products, and infrastructure.

Wang emphasized that "superintelligence is coming," and Meta must "take it seriously" by reorganizing to increase development speed. Most division heads will report directly to Wang, including Nat Friedman, former GitHub CEO, who leads product efforts.

The Vision Behind Meta’s New AI Structure

Meta’s reorganization centralizes efforts into distinct teams to focus on research, training, product, and infrastructure, aiming to build a world-class organization to reach superintelligence. Wang acknowledged that organizational changes can be "disruptive" but necessary to speed progress.

Key Leadership and Reporting Changes

Nat Friedman leads the product-focused team and reports directly to Wang. FAIR (Facebook AI Research) remains led by Rob Fergus, with Yann LeCun continuing as FAIR’s chief scientist; both report to Wang. Shengjia Zhao, co-creator of ChatGPT, leads MSL’s research efforts but is the only leader Wang did not mention as reporting directly to him.

Centralizing AI Research: The Roles of TBD Lab and FAIR

Research at Meta is now split between two units: TBD Lab and FAIR. TBD Lab is a small team concentrating on training and scaling large AI models, including exploring a new “omni” model. Wang’s memo does not detail what “omni” entails.

TBD Lab’s Focus on Large Model Training and the “Omni” Model

TBD Lab aims to achieve superintelligence across pre-training, reasoning, and post-training phases. The "omni" model may align with Meta’s focus on multimodal AI, as early hires included experts in audio, video, and other mediums.

FAIR’s Renewed Role as Innovation Engine

FAIR will play a more active role by feeding research directly into TBD Lab’s model training. Known for high-level AI research and relative independence, FAIR’s integration represents a shift towards collaboration within MSL.

Product Development and Infrastructure: Building Meta’s AI Future

Meta’s applied research and product development fall under Nat Friedman, who manages teams working on AI features such as Assistant, Voice, and Media pillars. Aparna Ramani leads the unified infrastructure team, responsible for building advanced systems to support AI research and products.

Nat Friedman Leads Product Integration Efforts

Friedman’s team brings product-focused research closer to product development, reporting to Wang. Meta has been trying to commercialize AI glasses and VR devices, but these have not yet become significant revenue sources.

Aparna Ramani, Heads of Unified AI Infrastructure Team

The infrastructure team consolidates multiple units to focus on accelerating AI research and production, building optimized GPU clusters, and developing comprehensive developer tools.

Dissolution of AGI Foundations and Team Realignment

Meta is dissolving the AGI Foundations team, created in May, redistributing members across product, infrastructure, and FAIR teams. This follows the prior dissolution of the GenAI division that developed Llama models.

Impact on AI Teams and Talent Redistribution

The restructuring moves AGI Foundations personnel to better-aligned teams. TBD Lab is not a destination for these members. Wang emphasized that while changes can be disruptive, the new structure will help reach superintelligence faster.

Meta’s AI Strategy in a Competitive Landscape

Meta’s frequent organizational changes contrast with rivals like OpenAI, Google, and Anthropic, which have experienced fewer upheavals. This raises questions about whether the new reorganization will restore Meta’s edge in AI development.

PHOTO: WEB SUMMIT YOUTUBE VIA TECHCRUNCH

This article was created with AI assistance.