Why AMI Labs’ CEO avoids calling his AI ‘AGI’

While much of the artificial intelligence sector rushes to brand its latest breakthroughs with flashy labels like AGI or superintelligence, the CEO of Yann LeCun’s world model startup, AMI Labs, is deliberately steering clear of the hype. Alexandre LeBrun recently told TechCrunch that his company has never used the term AGI, and he’s not impressed with its successor either.

“We never used the word AGI. And I just noticed that nobody is using it anymore; they switched to superintelligence,” LeBrun said. “Next time we’ll switch to something else.” He dismissed the new label as equally hollow, adding, “There’s no good definition. What is superintelligence? I don’t know. It’s not a very useful word.”

That is a pointed stance from a founder at the center of AI’s newest frontier. TechCrunch spoke with LeBrun last week in Seoul, where he was attending The International Conference on Machine Learning and scouting for local industrial partners, global companies, and researchers. AMI Labs is still pre-product, but it is already courting players in robotics, manufacturing, and electronics. LeBrun explained that a world model, which uses physics to predict and interact with the physical world, must prove its value beyond the lab.

One of the most promising applications for world models is robotics. Currently, robots operate on fixed routines, “completely static,” and AI remains “really dumb in the physical world,” LeBrun said. Even achieving basic context awareness in robots would mark “a very big difference for the world.” He cited a real-world example where a robot tasked with dancing and performing kung fu at a public event approached and kicked a child. Context-aware AI could have prevented that. “The hardware is very advanced; progress in hardware in the last few months is incredible, but there’s no brain.”

LeBrun drew a clear distinction between large language models (LLMs) and world models. An LLM predicts the next word or text, while a world model predicts the next state of the world. Nudge a glass off the table, and you intuitively know it will tip and spill. That instinct is what a world model aims to capture. However, he is not claiming world models are superior. LLMs are “complementary, not replaceable” for AI systems that need to understand the physical world. He compared the relationship to the human brain’s separate language and reasoning functions, arguing that LLMs will remain the most efficient tools for processing language, while world models will provide real-world context and understanding.

LeBrun believes nearly every industry that “touches the real world” could eventually benefit from robotics powered by world models. Physical environments, he argued, remain the biggest weakness for LLMs. A factory robot repeating the same motion works fine today, but the real challenge emerges when “you take your robot outside into a more open environment, in your household, or in the street.” There, the robot must understand its surroundings and operate safely. “Robots are not safe right now,” he said. “There’s no solution for that today.”

Healthcare offers a personal example for LeBrun, whose previous company, Nabla, was an AI health startup. He compared today’s AI systems to a doctor trained only on textbooks without any residency experience. LLMs may be useful in medicine, he said, but they cover “only 1% of healthcare.” The rest depends on real-world experience that a world model can provide.

But building a world model cannot happen in isolation. To train on reality, AMI needs real environments and close partners. “We need access to the real world,” LeBrun said, and it is “easier for us to do that with partners.” That is part of what draws him toward Asia, where the robots, chips, and factories actually exist. He won’t spell out a full Asia strategy yet. “It’s too early,” he said. But South Korea holds particular appeal for two reasons. First, it has advanced industries in robotics, semiconductors, and manufacturing, sectors that the first wave of AI barely touched. Second, Korea has a track record of rapid adoption. “Korea was the fastest adopter of the internet 25 years ago,” LeBrun noted. That combination of a deep industrial base and willingness to embrace AI fast is “unique,” and the reason “we want to be here from day one.”

JP Lee, CEO of SBVA and one of AMI’s backers in Asia, echoed that sentiment. “I’ve been telling Alex and the team to come to Korea,” he told TechCrunch. Lee acknowledged that the Korean government has done “a tremendous job” funding local sovereign LLM models, which already work “well enough” for general-purpose tasks. But he is pushing for Korea to keep investing in physical AI as well. He pointed to Seoul’s June plan to mobilize approximately $880 billion for chips, AI data centers, and physical AI, one of three declared pillars. “They should coexist,” Lee said. He also argued that Korea’s value to foreign firms extends beyond hardware, as local developers are quick to adopt and adapt new tools, a pattern that produced homegrown internet giants like Naver and Kakao.

For all the star power and the billion-dollar check, AMI Labs still has nothing to sell. The startup, co-founded by Turing Award winner Yann LeCun after he left Meta, raised $1.03 billion in March at a $3.5 billion pre-money valuation. There is no product yet, and no timeline LeBrun will commit to. “We’ll make a surprise when we’re ready,” he said.