The company has 12 employees and aims to develop intelligent systems that understand reality and operate safely in the real world.
Yann LeCun, a French-American computer scientist known for his contributions to the development of deep learning and former head of AI at Meta, has founded the startup Advanced Machine Intelligence (Ami), headquartered in Paris with additional offices in New York, Montreal, and Singapore. The company has 12 employees and aims to develop intelligent systems that understand reality and act safely in the real world, going beyond traditional language models like ChatGPT. Unlike LLMs that only predict the next words in text, Ami uses “world models” and the Jepa architecture to create abstract representations of the world that can predict the evolution of situations and make decisions based on certain objectives.
LeCun explains that this approach allows the system to avoid unpredictable details, learn general structures of reality, and plan realistic actions, a step that large language models cannot accomplish, despite their success in coding, mathematics, and writing.
Advanced Machine Intelligence plans to move in three phases: first research and development to consolidate its methodology and architecture, then collaboration with industrial partners in sectors such as automotive, aeronautics, biomedicine and pharmaceuticals to apply the technology to real-world cases, and finally the creation of a universal intelligent system. Among the founders and key leaders are Alexandre LeBrun as CEO, Saining Xie from Google DeepMind as Chief Science Officer, Pascale Fung as Chief Research and former Meta employees such as Michael Rabbat and Laurent Solly, while one of the potential customers could also be Meta for the use of the technology in Ray-Ban Meta smart glasses.
Investors include Cathay Innovation, Greycroft, Hiro Capital, HV Capital, Bezos Expeditions and Nvidia, and the company could be valued at $3.5 billion. LeCun says the project aims to change the way AI understands and acts in the real world, focusing on simulating situations and actions, rather than simply imitating language.