Why the AMI Labs Founder Refuses to Call His AI 'AGI' or 'Superintelligence'

When Terminology Becomes Mere Noise

At a time when AI companies are racing to claim the most sensational titles, Alexandre Lebrun, CEO of AMI Labs, is taking a decidedly different path. While competitors throw around terms like "artificial general intelligence" and "superintelligence," Lebrun prefers silence on these labels, viewing them as more problematic than valuable.

Speaking with TechCrunch on the sidelines of the International Conference on Machine Learning in Seoul, Lebrun said: "We never used the term AGI. And now everyone is dropping it in favor of superintelligence, which will itself be replaced by something else soon." He added plainly: "There's no precise definition of superintelligence, I don't know exactly what it means, and it's a term that adds no real value."

World Models: Beyond Large Language Models

AMI Labs' vision centers on what are known as "world models" — AI systems that integrate the laws of physics and physical logic to understand the surrounding environment and predict its future states. The distinction is fundamental: a large language model predicts the next word in a text, while a world model predicts the next state in actual reality.

Lebrun illustrates the idea with a simple example: when you push a glass toward the edge of a table, you instantly know it will fall and shatter. That physical intuition is precisely what world models aim to simulate and internalize. Yet Lebrun does not see these models as competitors to language models — rather as complements to them. He invokes a striking analogy: just as the human brain has separate centers for language and logical reasoning, language models will remain the most efficient tool for processing text, while world models will provide contextual understanding of the physical environment.

Robots: Bodies Without Minds

Lebrun pays close attention to the robotics sector, describing it as the field that most clearly exposes the weaknesses of current AI. Today's robots operate on fixed, repetitive programs, incapable of grasping context or adapting to changing variables. He offers a stark example: a robot programmed for dancing and kung fu moves at a public event ends up kicking a child, simply because it lacks sufficient environmental awareness to read the situation.

"The hardware progress over the past few months has been remarkable," Lebrun says, "but there is no mind." He believes that simply making robots "context-aware" would make an enormous difference — especially in open environments like homes and streets, where situations are unpredictable and safety becomes a non-negotiable requirement.

The Healthcare Experience: A Doctor Without Clinical Training

Lebrun draws on his previous experience with Nabla, a medical AI startup, to paint a telling picture of current systems' limitations: "Large language models are like a doctor who received a thorough textbook education but never completed any hands-on hospital training." He estimates that these models cover only one percent of actual healthcare needs, with the vast majority remaining dependent on field expertise and real-world interaction.

South Korea: A Strategic Partner, Not Just a Market

Lebrun's presence in Seoul was no passing visit — he is actively seeking local industrial partners in the robotics, semiconductor, and manufacturing sectors. He sees South Korea as a strategic destination for two primary reasons:

• Advanced industrial base: South Korea possesses a well-established infrastructure in manufacturing, electronics, and robotics — precisely the domains where world models need to prove their worth outside the laboratory.
• Speed of adoption and adaptation: Lebrun recalls South Korea's experience with the internet a quarter century ago, when it was among the world's fastest countries to adopt and leverage it. He believes this combination of industrial strength and rapid adoption capacity makes South Korea "a unique case" worth engaging with from day one.

This direction is backed by JP Lee, CEO of SBVA and one of AMI's Asian investors, who stresses that South Korea needs to invest in physical AI alongside language models, pointing to Seoul's plan to inject approximately $880 billion into processing chips, data centers, and embodied AI.

One Billion Dollars and Still No Product

AMI Labs — co-founded with Turing Award laureate Yann LeCun following his departure from Meta — raised $1.03 billion in funding in March 2024, at an initial valuation of $3.5 billion. Yet the company remains in pre-launch mode, with no product ready and no announced timeline. Lebrun sums up the situation in a single sentence: "We will surprise everyone when we are ready."

In a landscape where companies race to announce "revolutions" that have yet to materialize, Lebrun's stance seems rare: keep secrets until the right moment, and avoid inflated labels until systems prove their actual value in the real world.

✦ بقلم فريق دروب أيديا