Meta’s Chief AI Scientist Predicts Large Language Models Will Never Achieve Human Intelligence

LeCun pointed out that these models lack a fundamental understanding of the physical world

In a recent interview with the Financial Times, Meta’s chief AI scientist, Yann LeCun, expressed skepticism about the future capabilities of large language models (LLMs) such as ChatGPT, asserting they will never reach human-level intelligence. LeCun pointed out that these models lack a fundamental understanding of the physical world, do not possess persistent memory, and are unable to reason or plan in a meaningful way. LeCun emphasized the intrinsic limitations and safety concerns of relying on LLMs for achieving human-level cognition. He argued that these models depend heavily on the right training data to produce accurate responses, which inherently makes them unreliable and unsafe. Instead, LeCun is dedicating his efforts to developing a new generation of AI systems aimed at attaining human-level intelligence, though he acknowledges this endeavor might take up to a decade.

The pursuit of advanced AI is not without its challenges. Meta recently experienced a significant $200 billion drop in value after CEO Mark Zuckerberg announced plans to increase spending to establish the company as a leader in AI. This strategy highlights the tension between long-term AI development and investors' expectations for quick returns. Despite these challenges, other companies continue to push forward with enhanced LLMs. AI firm Scale recently secured $1 billion in a Series F funding round, with founder Alexandr Wang expressing ambitions for artificial general intelligence (AGI). Similarly, French startup H raised $220 million, with CEO Charles Kantor revealing their goal of achieving "full-AGI." However, the feasibility of AI reaching human-like thinking remains a topic of debate among experts. Akli Adjaoute, a veteran in the AI field, underscores the importance of recognizing AI's reliance on data and its inherent limitations. Adjaoute suggests that instead of speculating on AI's potential to replicate human intelligence, the focus should be on understanding its role as a powerful, yet fundamentally different, tool.