Behind the Scenes at AMI Labs: LeCun's 'World Model' Goal

The new venture AMI Labs, which renowned AI figure Yann LeCun left Meta to work on, has been a subject of curiosity since its inception. The company finally clarified its goals this week – although some significant clues were already apparent.

On its newly launched website, the company announced it will develop "world models" to build "intelligent systems that understand the real world." The acronym AMI, which stands for "Advanced Machine Intelligence," already hinted at this focus, but now it has officially joined the ranks of the hottest AI research startups.

Models That Connect with the Real World

Developing foundational models that bridge the gap between AI and the real world has become one of the field's most exciting pursuits. Even without a concrete product yet, it attracts top scientists and deep-pocketed investors. Direct competitor World Labs, founded by AI pioneer Fei-Fei Li, reached unicorn status shortly after emerging from stealth mode.

World Labs, which launched its first product Marble that generates physically consistent 3D worlds, is reportedly in talks for new funding at a $5 billion valuation. It's no surprise that investors want to invest in LeCun with the same enthusiasm. This is precisely why rumors are swirling that AMI Labs is raising funds at a $3.5 billion valuation.

So who might these investors be? According to Bloomberg, venture capital firms in talks with the startup include Cathay Innovation, Greycroft, and Hiro Capital, where LeCun serves as an advisor. Other potential investors mentioned include 20VC, Bpifrance, Daphni, and HV Capital.

An Unexpected Name in the CEO Chair

Whoever signs the checks, investors shouldn't overlook an important detail: as LeCun himself clearly stated, he is AMI's Chief Scientific Officer, not its CEO. That role belongs to Alex LeBrun, the co-founder and former CEO of Nabla, a healthcare AI startup with offices in Paris and New York.

LeBrun's move from Nabla to AMI is actually part of a partnership announced by Nabla last December. Nabla, which develops AI assistants for clinical care and where LeCun serves as an advisor, supported LeBrun's transition from CEO to Chief AI Scientist and Chairman of the Board in exchange for "privileged access" to AMI's world models. This move paved the way for his new role.

As CEO of AMI Labs, Alex LeBrun will be surrounded by familiar faces. After his previous startup Wit.ai was acquired by Facebook, this serial entrepreneur and AI engineer worked at Meta's FAIR AI research lab under LeCun's leadership. Reports indicate that Laurent Solly, who left his role as Meta's Vice President for Europe last December, will also join this duo.

The talent crossover between AMI and Meta likely won't end here. LeCun told MIT Technology Review that his former employer could be AMI's first customer. However, he has also publicly criticized some strategic choices made under Mark Zuckerberg's leadership.

A Bet Against Large Language Models

From a broader perspective, MIT Technology Review interprets AMI Labs as a different bet against large language models (LLMs). Among the LLM limitations pointed out by LeCun are hallucinations (generating imaginary answers), which is a serious concern in fields like medicine. LeBrun also knows this firsthand from his experiences at Nabla.

Imagine an AI model suggests a treatment to you, but this recommendation doesn't align with real physical laws or biological limits. The 'world models' that AMI Labs is pursuing seem precisely aimed at filling this gap. A system that understands and predicts the physical rules and causal relationships of the real world...

What does this mean? Future AI systems won't just generate text; they will be able to simulate the direction a ball will roll, how a complex mechanism will work, or how a drug will travel through the body. That fine line between understanding and simulating.

Although everything is currently in the research phase, these high-value funding talks happening in the background are an indicator of how much hope the industry places in this area. Having a name like LeCun at the helm of the company plays a critical role in both talent attraction and investor confidence. However, what's noteworthy is this: he is not taking on an operational role like CEO. This might stem from his desire to focus entirely on the scientific vision.

Ultimately, the next phase of the AI race promises to go far beyond language models. Companies like AMI Labs are trying to teach machines not just to 'talk,' but to 'understand' the real world. If they succeed, they have the potential to revolutionize countless fields, not limited to software but extending from robotics and autonomous vehicles to medical diagnosis and climate modeling. They are at the beginning of the road, but every step they take seems likely to leave a significant mark on shaping the future.