META’s chief AI scientist, Yann Lecun, says that the “new AI architecture new paradigm”, which far exceeds the abilities of the existing AI system, will appear within the next three to five years.
Lecun also predicted that the next few years, with the progress of AI and robot engineering, will be the “10 years of robot engineering”, where new class intelligent applications are released.
Lukan gave a lecture on Thursday in a session named “debate technology” at the Davos Conference, and at this time, the “AI taste”, that is, the generated AI and the large language model (LLM), in fact. He said that everyone was not satisfied. That much. It is certainly convenient, but not enough in many aspects.
“I think the current (LLM) paradigm’s life is quite short, probably three to five years,” Lucan said. “Within 5 years, at least as a central component of the AI system, I don’t think the person with sane will no longer use them. I think it will appear.
Lukan says these “restrictions” hinder the truly intelligent movement of the machine. There are four main reasons for this. Lack of physical understanding. Lack of permanent memory. Lack of inference. And lack of complex planning skills.
“LLM can’t actually do this at all,” says Lecun. “In other words, a more revolution of AI will occur in the next few years. Perhaps it is not generated in the sense we understand today, so it may be necessary to change the name.”
“World model”
This reflects the emotions that Lukan has supported in the past. At the center of this is the “world model” that machines are expected to help understand the real world dynamics. This includes the characteristics of the current system mainly, mainly for pattern recognition, such as memory, common sense, intuition, and inference abilities.
Lukan had previously stated that this could be about 10 years ahead, but today’s estimation has been close. However, it is not clear how much it will reach within that period.
“LLM is good at manipulating the language, but I’m not good at thinking,” says Lukan. “So what we are working on is to build the system of the world’s mental models. If the plan we are working on succeeded and we are on the schedule we want, 3-5 A completely different paradigm system will be born, and the world will probably work with the world. You may be able to learn.
“10 years of robot engineering”
The AI that can pass the bar exam or discover new drugs is wonderful, but Lukan can make robot engineering a central component of the next waves of AI applications in such real world scenarios. I think there is sex.
META itself conducts research activities in the field of robot engineering, but the same is true for Chatgpt creators, the most important person of the current AI. Earlier this month, new jobs that describe the details of the new Openai robot engineering team focusing on robots of “general -purpose”, “adaptability”, and “versatility” that can demonstrate human -like intelligence in the real world environment. Was announced.
“We don’t have a robot that can be made the same as cats. Understanding the physical world of the cat is much better than everything we can with AI,” he said. “Probably 10 years will be 10 years of robot engineering. Perhaps a wise AI system will appear to understand how the real world works.”
TechCrunch has a newsletter focusing on AI. If you sign up from here, you will reach the receiving box every Wednesday.