At the recent "Technology Debate" in Davos, Meta's Chief AI Scientist Yann LeCun made an exciting prediction about the development of artificial intelligence over the next five years. He believes that existing AI systems will undergo significant transformations within the next 3 to 5 years, leading to a "new AI architectural paradigm" that surpasses the capabilities of today's widely used generative AI and large language models (LLMs).
LeCun pointed out that while current LLMs perform well in language processing, they have significant limitations in truly intelligent behavior. He identified these limitations mainly in four areas: a lack of understanding of the physical world, a lack of persistent memory, a lack of reasoning ability, and a lack of complex planning capabilities.
Image Source Note: Image generated by AI, licensed from Midjourney
He candidly stated, "LLMs are not great at thinking, so we need to build systems with a better 'world model,' which will give machines common sense, intuition, and reasoning abilities."
LeCun emphasized that the new AI systems will learn through observation and interaction with the world, meaning that future machines will not only process information but will also better understand and respond to the complexities of the real world. He added, "If our plans can be successfully implemented, we will see entirely new AI systems emerge within the next 3 to 5 years, and these systems may possess some degree of common sense."
Meanwhile, LeCun also predicted that the next decade could be the "Decade of Robotics." He believes that the combination of AI and robotics will open up new intelligent application scenarios. He mentioned that although current generative AI is impressive, robots still have significant potential for development in understanding and adapting to the physical world. Current robotic technology is far from achieving the understanding capabilities of a cat.
Currently, Meta has already ventured into robotics research, and OpenAI is also actively investing in this field, forming a team focused on creating general-purpose, adaptable robots with the goal of developing robots with human-like intelligence. LeCun concluded that future AI systems will be smarter and are expected to truly understand how the real world operates.
