Yann LeCun - the world's leading scientist in the field of artificial intelligence, used to hold the position of Director of AI science at Meta, and was also the founder and leader of the AI FAIR research department of this group. After leaving Meta, Yann LeCun launched a startup company called AMI Labs.
On the official website, AMI Labs said it is developing world models, aiming to build an artificial intelligence system capable of understanding and interacting with the real world.
AMI is an abbreviation for Advanced Machine Intelligence. This goal brings AMI Labs into the group of AI startups pursuing a different direction from popular major language models today. Instead of focusing on language processing, LeCun believes that true intelligence must start from understanding the physical world, through sensor data, actions and the real environment.
The trend of developing world models is attracting great attention from investors. World Labs - a company co-founded by Fei-Fei Li, has quickly become a unicorn and is said to be negotiating to raise capital with a valuation of about 5 billion USD. Meanwhile, according to Bloomberg, AMI Labs is also being approached by venture capital funds, with a speculated valuation of about 3.5 billion USD.
Yann LeCun holds the role of CEO at AMI Labs, and the CEO is Alex LeBrun - co-founder of Nabla (an AI company in the medical field). LeBrun's move to AMI is part of a cooperation agreement between the two companies, to help Nabla access AMI's new generation AI models. LeCun also said Meta could become the first customer of the new company, although he has publicly criticized some of the AI strategies of this group.
According to LeCun and LeBrun, major language models today still have limitations such as "visional" phenomena, causing risks in areas requiring high accuracy, especially medicine. AMI Labs expects its models to have sustainable memory, reasoning, planning and ensuring safety when deployed in practice.
The company plans to license technology to industrial partners, and at the same time publish research and open source to cooperate with the academic community.
