In a recent celebration of the 10-year anniversary of Meta’s Fundamental Artificial Intelligence Research (FAIR) team, Yann LeCun, Meta’s AI boss, shared insights on various aspects of the artificial intelligence (AI) landscape.
LeCun, known for his cautious stance on achieving human-level AI, reiterated the need for incremental progress, suggesting that achieving “dog-” and “cat-” level AI is a prerequisite. In contrast, Elon Musk boldly predicted the arrival of a “digital god” within three to five years, while Nvidia CEO Jensen Huang offered a more conservative estimate of AI completing tests competitively with humans in the next five years.
LeCun, however, took a swipe at Huang, stating, “There’s an AI war, and he’s supplying the weapons,” alluding to Nvidia’s recent ascent as the world’s most valuable chip manufacturer. Nvidia’s GPUs have become the industry standard for training large language models like ChatGPT.
Addressing generative AI as a potential spark for Artificial General Intelligence (AGI), LeCun expressed skepticism, noting that the technology’s current state, particularly in understanding text, is lacking. Despite training systems on vast amounts of reading material, he emphasized the persistent challenge of comprehending simple logical relationships.
On the front of quantum computing, LeCun revealed Meta’s divergence from the quantum race, contrasting with competitors like Google and Microsoft. While acknowledging the scientific fascination of quantum computing, LeCun asserted that the technology is not yet mature, suggesting that classical computers remain more efficient for solving a broader range of problems.
In a landscape where Microsoft has invested $100 million in a partnership with Canadian quantum computing firm Photonic to develop a fault-tolerant quantum networking system within five years, Meta seems content to steer clear of the quantum computing fervor for the time being. LeCun’s perspective implies that practical applications for quantum computing are still limited compared to classical alternatives.