The Challenge to Superintelligence

meta CEO Mark Zuckerberg has invested billions in developing artificial intelligence technology with the goal of achieving superintelligence. Though, recent reports indicate critically important internal challenges are hindering progress, leading to a restructuring of Meta’s AI divisions.

Internal Disputes and Division Restructuring

According to a report by the New York Times, a dispute has emerged between veteran AI experts and newer hires within Meta. As a result, Meta Superintelligence Labs will be divided into four separate units: AI Research, AI Superintelligence, Products and Infrastructure, and AI Hardware. The company is also considering streamlining the overall AI division, perhaps through layoffs or reassignment of personnel, as it has grown to encompass thousands of employees.

Adding to the complexity, Meta is reportedly exploring the possibility of purchasing AI models from third-party developers to supplement its own internal efforts, as noted by the New York Times.

Significant Investment and Recent Hiring

In June, meta invested $14.3 billion in AI startup Scale AI and appointed its CEO, Alexandr Wang, as Meta’s head of AI. The company has also been aggressively recruiting top AI talent, offering substantial compensation packages. For example, 24-year-old AI expert Matt Deitke was reportedly offered a $250 million contract to join Meta.

meta’s total capital expenditure is projected to reach $72 billion this year, with a substantial portion allocated to AI development, according to the New York Times.

Wang’s Vision and Resulting tension

The restructuring is largely attributed to tensions stemming from Alexandr Wang’s vision for Meta’s AI strategy.PCMag reports that Wang intends to pursue a closed-source approach to Meta’s superintelligent AI model, rather then an open-source one.

This reportedly led to the abandonment of an existing superintelligence model, known as Behemoth, and the initiation of a new project from scratch. This decision reportedly prompted some long-term AI employees to leave the company due to disagreements with Wang’s approach.

Zuckerberg’s Optimism and the Path to ASI

Previously, Meta researchers claimed their AI systems were demonstrating the ability to self-improve without human intervention. Zuckerberg stated, “In the last few months, we began to glimpse that our AI systems were improving themselves. The development was indeed slow for now, but it was undeniable,” as reported by various sources including Meta’s official newsroom.

Zuckerberg views this self-improvement capability as a crucial step towards achieving artificial superintelligence (ASI), defined as AI that surpasses human cognitive abilities.