AI Pioneer Yann lecun Urges Shift Beyond Large Language Models for True Artificial General Intelligence
To achieve human-level AI, researchers must focus on embodied learning and a deeper understanding of the physical world, according to leading AI scientist Yann LeCun.
Seoul, South Korea – In a keynote address at the ‘Global AI Frontier Symposium’ hosted by the ministry of Science and ICT, New York University Professor Yann LeCun – also Global AI Frontier Lab Co-Director and Meta Chief Scientist – argued that current approaches to artificial intelligence, especially those centered on Large Language Models (LLMs), are insufficient for achieving true Artificial General Intelligence (AGI). LeCun emphasized the critical need for AI systems to move beyond processing text and engage with the physical world in a more extensive way.
“AI is not as smart as a house cat in terms of understanding the physical world,” he remarked, drawing a parallel to the Morabel Paradox – the observation that tasks easy for humans are often difficult for computers, and vice versa. Bridging this gap, he asserted, is essential for realizing AGI.
While acknowledging the extraordinary data processing capabilities of AI – capable of absorbing facts equivalent to 400,000 to 500,000 years of human reading in a short timeframe – LeCun highlighted the qualitative difference between data exposure and genuine understanding. He pointed to the formative years of a human child, noting that a four-year-old accumulates more data through sensory experience than an LLM processes through text alone. “For the first four years after a baby is born, a baby is awake for nearly 16,000 hours and checks with 2 million nerves,” he explained. “In fact, a 4-year-old baby in a physical habitat secures more data than an LLM through the five senses.”
LeCun advocated for AI growth inspired by human learning, specifically the process of observing and interacting with the world. He suggested exploring approaches like real-time vision AI to enable AI to understand the physical world through direct observation. He proposed a ‘world model’ methodology – developing hypotheses, imagining potential actions and states, and refining understanding through repeated learning – as a pathway toward this goal.
Beyond the technical challenges, LeCun stressed the importance of a global, open-source AI ecosystem. He argued that AI is rapidly becoming a fundamental resource for all nations and industries, and that monopolization by a few entities would stifle innovation. “Open source has always been a great motivation for the development and advancement of software,” he predicted, noting China’s increasing focus on open-source AI models.
Other experts at the symposium echoed the call for a broader approach to AI development. Stanford University Professor Yejin Choi, a senior director at NVIDIA and a member of Time magazine’s “2025 World’s Top 100 AI People,” emphasized the need for differentiated data and algorithms to create competitive, lightweight language models (SLMs) that can democratize AI access. “AI competitiveness is about learning quickly by making good use of limited data,” Choi stated.
The symposium also addressed the critical issue of AI safety and reliability.Experts like Professor Joshua Bengio of the University of Montreal acknowledged AI’s potential for positive change but cautioned about the inherent risks of AGI. professor Geoffrey Hinton of the University of Toronto acknowledged the difficulty of slowing AI development globally, even with safety concerns, and underscored the need for research focused on ensuring human-AI coexistence.
The South Korean government signaled its commitment to advancing AI technology through the National AI Research Base – a collaboration between KAIST, Korea University, Yonsei University, and Pohang University of Science and Technology – and the Global AI Frontier Lab. Deputy Prime minister and Minister of Science and ICT, bae Kyung-hoon, affirmed the nation’s dedication to international collaboration and nurturing the next generation of AI leaders.”We are building a cooperation network with researchers around the world,” he said, “and will strengthen AI competitiveness based on international cooperation.”
