Yann Lecun, a meta chief AI scientist, states that “AI Architecture New Paradigm” has appeared in the next three to five years, far exceeding the functions of the existing AI system.
LECUN also predicted that the next few years could become a “10 -year robot engineering”, and combined with the progress of AI and robot engineering to unlock the new class intelligent application.
In a session called “debate technology” in Davos on Thursday, Lecun states that the “AI flavor” we have at this time, that is, the generated AI and the major language model (LLMS) are not really all. I did it. so. It is certainly useful, but in many aspects.
“The current (LLM) paradigm’s storage life is quite short, probably 3-5 years,” says Lecun. “Within five years, no one uses them in their right heart, not at least as the central component of the AI system. (…) However, this may not be limited to the current AI system. “
Lecun says that these “restrictions” hinder the true intelligent behavior of the machine. There are four important reasons for this. Lack of physical understanding. Lack of sustainable memory; lack of inference; lack of complex planning ability.
“LLMS really has no ability,” Lecun said. “So another revolution of AI will occur in the next few years, probably not generated in the sense that you understand it today.”
“World model”
This reflects the emotions supported by Lecun in the past. At the center of this, it is known as a “world model” that promises to help the machine understand the real world dynamics. This has memory, common sense, intuition, and inference ability. This is far beyond the characteristics of the current system, mainly related to pattern recognition.
Previously, Lecun states that this may be about 10 years away, but today’s estimates are approaching the horizon. However, it is not clear how much you reach in that time frame.
“LLM is good at language operation, but I don’t think,” says Lecun. “That’s what we are working on. The system is to build a mental model of the world. If the plan we are working on succeeded, we use the timetable we want, 3-5. There is a completely different paradigy system within the year, and they may probably have a certain common sense. You may be able to do it.
“10 years of robot engineering”
Generative AI is impressive to pass the BAR exam or discover new drugs. Lecun believes that robot engineering may be the central element of the next wave of AI applications in such real world scenarios.
The meta itself conducts some research work in the area of robot engineering, but is the instant AI darling, Chatgpt-Creator Openai. Earlier this month, a new employment list has appeared, and the new Openai Robotics team focuses on “general -purpose”, “adaptation”, and “versatile” robots, becomes a robot that can be human -like in the real world settings. I focused on it.
“There is no robot that can do what cats can do. Understanding the physical world of cats is much better than everything that AI can do,” he said. “Maybe the next 10 years will be 10 years of robot engineering, perhaps there is enough wise AI system to understand how the real world works.”
TechCrunch has a newsletter focusing on AI! Sign up here and get it on the reception tray every Wednesday.
