A New Open Source AI Giant Emerges: Zuckerberg Sets Sights on AGI

In an unexpected revelation poised to reshape the landscape of artificial intelligence, Meta CEO Mark Zuckerberg announced his ambitious goal to develop AI models that could potentially achieve artificial general intelligence (AGI). This initiative positions Meta as a significant contender in the open-source AI arena, setting it apart from proprietary models developed by companies like OpenAI.

Zuckerberg articulated the vision for AGI in a recent Facebook post, stating, “The next generation of services necessitates the creation of full general intelligence—advanced AI assistants, tools for creators, business solutions, and much more.” He underscored the need for advancements across various AI domains, including reasoning, planning, programming, memory, and other cognitive functions. “Given the critical importance of this technology and the vast opportunities it presents, we are committed to open-sourcing it to the extent that is responsible,” he added.

The concept of general intelligence, fundamentally understood as the capacity for AI systems to perform tasks at or above human capability, has been a long-standing aspiration for leading AI organizations, including OpenAI and Google DeepMind. However, as Zuckerberg noted, there is no universally accepted definition of AGI. He explained, "I cannot provide a simple definition; the term encompasses various interpretations—ranging from human-level intelligence to enhanced, future forms of intelligence. What matters most is the extensive range of capabilities inherent in intelligence, which includes reasoning and intuition.”

This pursuit of AGI also serves as an attractive incentive for top-tier talent, as Zuckerberg mentioned, “Many of the brightest researchers are drawn to tackle the most ambitious challenges.” With this renewed focus, Meta aims to build on its existing AI projects, such as the LLaMA and its successor, Llama 2—foundational models for many open-source AI systems. Zuckerberg disclosed that Meta is actively working on Llama 3, the next generation of its large language models.

However, he remained non-committal about whether an AGI breakthrough would be open-sourced. “As long as it is prudent and aligns with safety considerations, we generally aim for open-source. Yet, we cannot be bound to a promise that might not serve us well,” he stated.

To bolster its AI strategy, Meta is consolidating its Fundamental AI Research (FAIR) lab with its AI product division, GenAI. This strategic move will have Joelle Pineau, head of FAIR, and Yan LeCun, Meta’s chief AI scientist, report directly to Chris Cox, the company’s chief product officer.

Additionally, Zuckerberg announced plans for a substantial infrastructure upgrade to support its AI ambitions, including the procurement of 350,000 H100 chips from Nvidia by the end of 2024. This investment will nearly double Meta’s computing power, bringing it close to the equivalent of 600,000 H100 units. Research firm Omdia predicts that Meta will continue to be a leading buyer of Nvidia's AI chips, having already ordered at least 150,000 H100 chips in 2023 alone.

Zuckerberg also expressed skepticism toward Meta's competitors in the AI space, pointing out that many companies that once promoted transparency in their research are now opting for discretion. “It’s interesting to see how firms that previously embraced openness have shifted to valuing proprietary developments more significantly,” he remarked.

Contrary to the prevailing narrative, Zuckerberg clarified that Meta has not abandoned its metaverse aspirations despite its intensified focus on AI. In fact, he stated, “New devices will be essential for AI, merging the realms of AI and the metaverse. In the future, many of us will interact with AIs regularly throughout our day.” He highlighted smart glasses as the ideal medium for integrating AI technologies, enabling AIs to perceive our surroundings. This vision is already taking shape through Meta’s collaboration with Ray-Ban to create smart glasses equipped with a camera and an integrated AI assistant. Initial reception of these glasses has been promising.

LeCun reinforced the notion that future interactions in digital environments will be seamlessly facilitated by AI assistants through smart glasses and similar devices. “For these systems to assist us effectively in our daily lives, they must possess human-level intelligence to navigate and comprehend our world,” he concluded.