Yang Lucan leaves Meta to explore new directions in AI research

Renowned AI researcher Yan Lukan has announced his departure from Meta after 12 years and plans to establish an independent AI company. This decision is attracting attention as it suggests a major shift in the current direction of AI development.

Shifting Technical Strategies from ‘Irrational Attachment’ to LLMs

Industry reports in November 2024 revealed that Yan Lukan intends to leave Meta. He is raising fundamental questions about the current AI industry’s overemphasis on large language models (LLMs). In his view, the mainstream approach today is characterized by an “irrational attachment,” which he criticizes harshly.

Lukan has dedicated many years to building theoretical frameworks for AI evolution. Based on his experience, he argues that chatbot technology alone makes it difficult to realize AI systems that surpass human intelligence levels. While LLMs demonstrate excellent capabilities in language processing, he and other researchers see fundamental limitations, emphasizing the need for different approaches.

Why World Models Are Key to Achieving AGI

The new company Yan Lukan is establishing will focus on developing “world models,” an AI technology concept that involves integrating data from visual inputs, spatial information, and various sensors to deeply understand the laws and structures of the real world.

Lukan asserts that constructing cognition mechanisms based on the physical world is the essential path to achieving Artificial General Intelligence (AGI). He believes that learning through interaction with the environment, rather than just language processing, is crucial for reaching true intelligence levels.

AI Researchers Sharing the Same Vision

Interestingly, Yan Lukan’s perspective aligns with that of Fei-Fei Li, a former professor at Stanford University. Li also points out that the next frontier in AI development lies in “spatial intelligence,” and describes LLMs as “eloquent in expression but lacking in experiential understanding.”

Li’s analysis indicates that while LLMs excel at pattern recognition in language data, they still face challenges in understanding the essence of the world. This concern clearly echoes Lukan’s own issues. The fact that multiple leading researchers are pointing in the same direction around the same time suggests a significant turning point for the entire AI industry.

The shift toward world models and spatial intelligence, and the成果 that Yan Lukan’s new company aims to produce, are drawing close attention from the AI industry as a whole.