Alexandre LeBrun, CEO of AMI Labs, said the company does not use the terms “AGI” (artificial general intelligence) or “superintelligence.” In an interview with TechCrunch he said: “We never used the word AGI. And I just noticed that nobody is using it anymore; they switched to superintelligence. Next time we’ll switch to something else.” He added that such labels are poorly defined and not particularly useful.
Why AMI Labs is focusing on world models
AMI Labs is developing so‑called world models that incorporate physical reasoning to predict the next state of the real world. LeBrun contrasted this with large language models (LLMs): an LLM predicts the next word or piece of text, while a world model predicts the next state of the environment. His example: if you nudge a glass off a table, a world model should predict it will tip and spill.
Because world models must work outside controlled experiments, AMI Labs is seeking industrial partners in robotics, manufacturing and electronics. The company remains pre‑product but is already courting potential collaborators to provide access to real environments for training and evaluation.
Robotics and the challenges of the physical world
LeBrun expects world models to have particular impact in robotics. He argued that today most robots follow fixed, static routines and that “AI remains really dumb in the physical world.” Even giving robots basic context awareness, he said, would represent a significant advance — for example, preventing an incident in which a dancing robot at a public event moved toward and kicked a child.
He noted that hardware progress in recent months has been impressive, but that the “brain” — reliable decision making and safety in physical settings — is still missing.
LLMs and world models are complementary
LeBrun stressed he is not claiming world models are superior to LLMs; rather, they complement each other. Drawing an analogy to the human brain’s separate language and reasoning capacities, he said LLMs will remain the most efficient tools for language processing, while world models will provide context and real‑world understanding.
Nearly every industry that “touches the real world” could eventually benefit from robotics based on world models, LeBrun argued. While a factory robot repeating the same motion works adequately today, the challenge arises when robots operate in more open environments — households or streets — where they must understand surroundings and be safe. “Robots are not safe right now,” he said. “There’s no solution for that today.”
Healthcare and the need for real‑world experience
LeBrun pointed to healthcare as a domain where real‑world experience is critical. His previous company was the AI health startup Nabla. He compared current AI systems to a doctor trained only on textbooks and without residency experience. While LLMs can be useful in medicine, he said, they currently cover only “1% of healthcare”; the remainder relies on practical, real‑world knowledge.
Real‑world access and a pull toward Asia
According to LeBrun, a world model cannot be built solely in a lab: AMI needs access to real environments and close partners for training. This necessity is one reason the company is looking to Asia, where robots, chips and factories are physically concentrated.
LeBrun would not lay out a full Asia strategy yet, saying it is “too early,” but he cited two reasons South Korea is attractive: a strong industrial base in robotics, semiconductors and manufacturing, and speed — a national willingness to adopt new technologies quickly. He noted Korea’s track record as an early adopter of the internet 25 years ago and said the combination of deep industry and rapid adoption is “unique” and why AMI wants to be in Korea from day one.
JP Lee, CEO of SBVA and one of AMI’s backers in Asia, told TechCrunch he has encouraged the AMI team to come to Korea. Lee praised the government’s work in funding local sovereign LLMs, which he said already perform “well enough” for general tasks, but he is pushing for continued investment in physical AI. He pointed to Seoul’s June plan to mobilize about $880 billion for chips, AI data centers and physical AI as evidence that the three should coexist.
Lee also argued Korea’s value to foreign firms is not only in hardware; local developers quickly adopt and adapt new tools, a pattern that produced domestic internet companies such as Naver and Kakao.
Funding and product timeline
AMI Labs was co‑founded by Turing Award winner Yann LeCun after he left Meta. The startup raised $1.03 billion in March at a $3.5 billion pre‑money valuation. Despite the large funding round and high‑profile founders and backers, AMI currently has no product on the market and LeBrun would not commit to a timeline. “We’ll make a surprise when we’re ready,” he said.
Conclusion
AMI Labs is avoiding grandiose labels and focusing on building world models that learn from real‑world interactions. The company is actively seeking partners in robotics, manufacturing and electronics to move models out of the lab and into physical environments, with particular interest in South Korea because of its industrial base and speed of adoption.
