Mark Zuckerberg has propelled Meta, the tech giant formerly known as Facebook, into the competitive landscape of Artificial General Intelligence (AGI) development. In a bold move contrasting with industry peers, Zuckerberg has committed to an open-source future for Meta’s AGI technology. This strategy diverges from the paths taken by Google and OpenAI, which have opted for proprietary developments amid concerns about the safety of open-sourcing AI.
Meta’s Shift in Strategy
Historically, Meta has maintained a product-centric AI focus. However, Zuckerberg’s latest announcement heralds a significant pivot towards a more generalized AI approach. This transformation is underscored by Meta’s past VP of AI, Jerome Pesenti, who once criticized AGI as overhyped. Yet now, under Zuckerberg’s directive, Meta is aligning its AI endeavors towards the creation of AGI.
Meta’s existing AI models, such as Llama 2, have been shared openly, and Zuckerberg affirms this philosophy will persist into the AGI era. The implications of open-source AGI raise questions about the feasibility of safe and responsible dissemination—a challenge Zuckerberg and Meta’s chief AI scientist, Yann LeCun, believe can be met.
Consolidation and Resources
In pursuit of AGI, Meta is restructuring, merging its Fundamental AI Research (FAIR) division with the GenAI team responsible for app development. This consolidation is driven by the recognition that AGI will be central to the next wave of digital assistants and services, necessitating close collaboration between application developers and AI model researchers.
Like its tech counterparts, Meta is contending with a talent crunch. By uniting its AI divisions, Meta aims to streamline its focus and resources towards AGI—a key strategy to outpace Google and OpenAI in the AGI race.
Zuckerberg has also revealed Meta’s investment in a vast infrastructure expansion, with projections to possess approximately 350,000 NVIDIA H100 GPUs by the end of 2024. This will complement Meta’s current computing capabilities, effectively equating to the power of 600,000 H100 GPUs. These resources are currently being channeled into training Llama 3, Meta’s next-generation AI model, though its release date remains undisclosed.
Hardware and Adoption
Meta’s commitment to open-source extends beyond software; it’s a strategic component of their hardware adoption goals as well. Zuckerberg continues to prioritize the Metaverse, envisioning a future where new devices will leverage AI advancements. To this end, Meta promotes a lineup of devices, including Ray-Ban Meta Glasses and Quest VR goggles, as integral to experiencing the Metaverse.
The exact definition of AGI remains nebulous within the industry. Despite this, there’s anticipation that AGI will exhibit intelligence at or beyond human levels. LeCun, however, tempers expectations, suggesting that “cat-level” or “dog-level” AI will precede any form of human-equivalent AI. As the industry inches closer to AGI, the hope is that the potential risks will be managed as diligently as the technology itself.
English