While much of the artificial intelligence sector actively labels its advancements with terms like “AGI” or “superintelligence,” Alexandre LeBrun, CEO of AMI Labs—Yann LeCun’s world model startup—consciously steers clear of such descriptors. In a recent interview with TechCrunch, LeBrun affirmed that the company entirely avoids using phrases such as “AGI” or “superintelligence.”
“We never used the word AGI. And I just noticed that nobody is using it anymore; they switched to superintelligence,” he remarked, adding, “Next time we’ll switch to something else.” He expressed similar skepticism about the newer label. “There’s no good definition. What is superintelligence? I don’t know. It’s not a very useful word.”
This represents a deliberate stance from a founder positioned at the forefront of the AI industry’s latest competitive frontier.
TechCrunch engaged with LeBrun last week during his visit to Seoul for The International Conference on Machine Learning. His trip involved actively seeking out local industrial partners, global corporations, and researchers. Although AMI Labs is currently in its pre-product phase, it is already engaging with key players in robotics, manufacturing, and electronics. LeBrun emphasized that a world model, which integrates principles of physics to predict and interact with the physical world, must demonstrate its capabilities beyond laboratory settings.
One domain where world models are anticipated to exert substantial influence is robotics. LeBrun noted that contemporary robots are confined to executing fixed, “completely static” routines, and current AI remains “really dumb in the physical world.”
He articulated that even if AI could simply enable robots to become “aware of the context,” this would signify “a very big difference for the world.” Such context-aware AI could have been instrumental, for instance, in preventing a robot performing at a public event from inadvertently approaching and kicking a child. LeBrun highlighted, “The hardware is very advanced; progress in hardware in the last few months is incredible, but there’s no brain.”
Drawing a distinction, LeBrun explained that a large language model (LLM) forecasts the subsequent word or text, whereas a world model anticipates the next state of reality. He illustrated this with a common intuition: if you nudge a glass off a table, you instinctively know it will tip and spill—this is precisely the kind of predictive understanding a world model aims to encapsulate.
LeBrun clarified that he does not assert the superiority of world models over LLMs, stating they are “complementary, not replaceable” within AI systems designed to comprehend the physical world. He drew an analogy to the distinct language and reasoning functions of the human brain, suggesting that LLMs will remain the optimal tools for language processing, while world models will furnish context and real-world comprehension.
According to LeBrun, virtually every industry that “touches the real world” could eventually benefit from robotics powered by world models, particularly given that physical environments represent a significant weakness for LLMs.
He acknowledged that a factory robot performing repetitive motions functions adequately today. However, the true challenge emerges when “you take your robot outside into a more open environment, in your household, or in the street,” where it must interpret its surroundings and operate safely. LeBrun cautioned, “Robots are not safe right now,” adding, “There’s no solution for that today.”
For LeBrun, whose previous venture was the AI health startup Nabla, healthcare provides a more personal illustration. He likened present-day AI systems to a doctor educated solely through textbooks, lacking any practical residency experience. While LLMs may contribute to medicine, he argued they address “only 1% of healthcare,” with the remainder reliant on real-world practical experience.
However, LeBrun stressed that a world model cannot be developed exclusively within a lab. To effectively train on reality, AMI requires access to real-world environments and close collaboration with partners. “We need access to the real world,” he stated, and it’s “easier for us to do that with partners.” This imperative is a key factor drawing him towards Asia, a hub for robotics, chip manufacturing, and advanced factories.
LeBrun refrained from detailing a comprehensive Asia strategy, noting, “It’s too early.” Nevertheless, South Korea’s appeal stems from two primary factors. Firstly, Korea boasts advanced industries in robotics, semiconductors, and manufacturing—hardware-intensive sectors that were largely untouched by the initial wave of AI development.
The second compelling attraction is speed. LeBrun highlighted Korea’s national initiative to significantly invest in AI and its established history as an early adopter. “Korea was the fastest adopter of the internet 25 years ago,” he observed. This distinctive blend of a robust industrial base and a readiness to rapidly embrace AI is what he describes as “unique,” and the reason “we want to be here from day one.”
“I’ve been telling Alex and the team to come to Korea,” JP Lee, CEO of SBVA and one of AMI’s Asian investors, informed TechCrunch.
Lee commended the government’s “tremendous job” in funding local sovereign LLM models, which he believes already perform “well enough” for general tasks. However, he advocates for Korea to continue investing substantially in physical AI as well. He referenced Seoul’s June plan to allocate approximately $880 billion towards chips, AI data centers, and physical AI, designating it as one of three foundational pillars. “They should coexist,” Lee asserted.
Lee further argued that Korea’s value to international firms extends beyond its hardware capabilities. Local developers are quick to adopt and adapt new tools, a pattern that has fostered the growth of indigenous internet giants such as Naver and Kakao.
Despite its considerable star power and billion-dollar backing, AMI currently has no product available for sale. The startup, co-founded by Turing Award laureate Yann LeCun after his departure from Meta, successfully raised $1.03 billion in March, securing a pre-money valuation of $3.5 billion. With no product yet and no definitive timeline, LeBrun simply stated, “We’ll make a surprise when we’re ready.”
The Editorial Staff at AIChief is a team of professional content writers with extensive experience in AI and marketing. Founded in 2025, AIChief has quickly grown into the largest free AI resource hub in the industry.