Meta’s Chief AI Scientist Yann LeCun Questions the Longevity of Current GenAI and LLMs 

Source: ShutterStock

The current paradigm of generative AI (genAI) and large language models (LLMs) may soon be obsolete, according to Meta’s Chief AI Scientist, Yann LeCun. He argues that new breakthroughs are needed for the systems to understand and interact with the physical world. 

Speaking at the World Economic Forum in Davos, LeCun remarked on GenAI systems: “Nobody in their right mind would use them anymore, at least not as the central component of an AI system.” He expects a revolution will occur in the next three to five years during which a new paradigm of AI architectures will emerge that far exceeds the capabilities of current systems.

LeCun remarks are not necessarily a dismissal of the remarkable progress achieved with genAI and LLMs, but more a recognition of the limitations of the current systems. According to LeCun, the current LLMs lack a true understanding of the physical world, and there is a long way to go before AI can match humans in this aspect. 

Meta’s AI chief Yann LeCun

“There are still a lot of scientific and technological challenges ahead, and it’s very likely that there’s going to be yet another AI revolution over the next three to five years because of the limitations of current systems,” said LeCun. “If we want eventually to build things like domestic robots and completely autonomous cars, we need systems to understand the real world.”

Regarded as one of the “Godfathers of AI”, LeCun is of the opinion that while AI excels at tasks like language manipulation it falls short when it comes to understanding the real world. He predicts that the coming years will be a “decade of robotics,” where advances in AI and robotics combine to unlock a new class of intelligent applications. 

Along with limitations in understanding the physical world, LeCun feels that LLMs also lack persistent or continuous memory, reasoning capabilities, and the ability to perform complex planning tasks. LeCun's comments align with his past statements, where he argued that LLMs are not close to reaching human intelligence and may never reach that point. 

LeCun’s perspective is contrary to the growing hype around AGI and superintelligence. Sparked by breakthroughs like ChatGPT and agentic AI, many assume that we are on the brink of creating machines with superintelligence capabilities. However, LeCun's predictions are shared by many AI experts, who believe true AI must integrate physical and digital realms. 

LeCun and his team at Meta are working on developing AI systems that build mental models of the world. “If the plan that we're working on succeeds, with the timetable that we hope, within three to five years we'll have systems that are a completely different paradigm," shared LeCun. "They (new systems) may have some level of common sense. They may be able to learn how the world works from observing the world and maybe interacting with it."

LeCun is one of the seven engineers awarded the prestigious £500,000 Queen Elizabeth (QE) Prize for Engineering for their contributions to modern machine learning, which has fueled advances in AI. LeCun’s fellow QE winner, Yoshua Bengio, warned that more progress is needed on the safety of the technology. 

Bengio is urging industry leaders to prioritize AI safety at the two-day AI Action Summit in Paris, which is taking place this week. “I’d like to see the leaders of this world better understand the magnitude of what we are doing, both in terms of the power we’re creating, which could be for good or be dangerous, and the risks that come with that power,” said Bengio. 

The AI Summit is expected to be attended by representatives from nearly 100 countries including major players like the US, China, and India. A key point on the agenda is to find common ground for sustainable AI development while balancing national priorities. Additionally, discussions will focus on streamlining AI regulations, advancing clean energy strategies, and fostering regional self-reliance in the AI sector. 

Beyond these policy issues, technical debates are also set to take center stage. LeCun spoke at the Summit and recommended prioritizing joint-embedding architectures instead of GenAI models. He emphasized that if the goal is to create human-level AI, then LLMs are not the way to go. According to LeCun, AGI can only be achieved using World models like JEPA, and not LLMs.