Yann LeCun’s World Model Startup Won't Say AGI—or Superintelligence

Alexandre LeBrun is done with the buzzwords. The CEO of AMI Labs, the world model startup launched by Meta’s chief AI scientist Yann LeCun, is steering clear of the industry’s favorite labels. “We never used the word AGI. And I just noticed that nobody is using it anymore; they switched to superintelligence,” LeBrun told TechCrunch during a trip to Seoul. “Next time we’ll switch to something else.” His blunt dismissal of the term superintelligence is equally sharp: “What is superintelligence? I don’t know. It’s not a very useful word.” It’s a refreshingly honest stance from a founder at the center of a new AI race, one that prioritizes physics over hype.

That race is about building a brain for the physical world. LeBrun’s pre-product company is hunting for industrial partners in robotics, manufacturing, and electronics, arguing that current AI is “really dumb in the physical world.” Robots today run fixed routines and lack even basic context awareness. He offered a grim example: a dancing robot at a public event that kicked a child. “The hardware is very advanced; progress in hardware in the last few months is incredible, but there’s no brain,” he said. A world model, which predicts the next physical state of a glass tipping off a table rather than the next word in a sentence, is meant to fill that gap. LeBrun doesn’t see it as a replacement for large language models but as a complementary system, much like the human brain’s separate functions for language and reasoning.

The search for real-world training data is pulling AMI Labs toward Asia. LeBrun was in South Korea for the International Conference on Machine Learning, scouting partners who can offer what a lab cannot: actual factories, chips, and unpredictable environments. To train a model on reality, you need messy, physical inputs. South Korea’s attraction is twofold. First, its deep industrial base in robotics and semiconductors represents sectors the first wave of AI barely touched. Second, LeBrun pointed to the country’s cultural speed, noting Korea was the fastest adopter of the internet 25 years ago. “That combination, a deep industrial base plus a willingness to embrace AI fast, is unique,” he said. JP Lee, CEO of SBVA and an AMI backer, is pushing for investment in physical AI, pointing to Seoul’s plan to mobilize some $880 billion for chips and AI data centers.

LeBrun’s fixation on the physical world comes with a personal understanding of AI’s gaps. His previous company was the healthcare startup Nabla. He likened today’s AI to a doctor trained only on textbooks, useful but covering maybe 1% of healthcare. The remaining 99% depends on real-world experience that a world model could eventually provide. For now, there is no safe solution for a robot operating in a household or on a street. LeBrun’s refusal to traffic in labels like superintelligence isn’t just semantics. It’s a signal that his startup’s value won’t be proven in a press release, but in a factory where a machine finally understands its surroundings.