While other companies in the AI industry race to label their work “AGI” or “superintelligence,” Alexandre Leblanc, CEO of Yann LeCun’s global model startup AMI Labs, avoids the term altogether. Lebrun said in an interview with TechCrunch that the company doesn’t use terms like “AGI” or “superintelligence” at all.
“We never used the term AGI, and I just realized that no one uses it anymore. They switched to the term superintelligence,” he says. “Next time I’ll switch to something else.” He’s also not sold on the new label. “There’s no good definition. What is superintelligence? I don’t know. It’s not a very useful term.”
It’s a sharp stance from a founder sitting at the center of AI’s latest race.
TechCrunch spoke with LeBlanc last week when he was in Seoul for an international conference on machine learning, looking for local industrial partners, global companies and researchers. Although still in the pre-production stage, AMI Labs has already begun working with robotics, manufacturing, and electronics companies. LeBlanc explained that world models that incorporate physics to predict and manipulate the real world need to prove their existence outside of the laboratory.
One area where the world model is expected to have a major impact is robotics. For now, LeBlanc said, the robots are “completely static” and run routines, leaving the AI ”totally stupid in the physical world.”
Even if AI could simply make robots “context aware,” it would make “a huge difference to the world.” Such context-aware AI could, for example, help prevent a dancing or kung fu robot from approaching a child and kicking him at a public event. “The hardware is very advanced. The advances in hardware over the last few months have been amazing, but there’s no brains.”
Large-scale language models (LLMs) predict the next word or text, and world models predict the next state. We already know that if you gently push the glass off the table, it will tip and spill. That’s the intuition that world models are meant to capture, LeBlanc explained, to predict the next state of the world.
LeBlanc does not claim that world models are better than LLMs when it comes to AI systems that understand the physical world. He said the LLM is “complementary and cannot be replaced.” He drew parallels between the unique language and reasoning capabilities of the human brain, adding that LLM will continue to be the most efficient tool for processing language, while world models will provide context and understanding of the real world.
LeBlanc said nearly every industry that “touches the real world” could eventually benefit from robotics based on world models, arguing that the physical environment remains where LLMs are weakest.
Factory robots that perform the same actions over and over again are still functioning well, he said. The challenge begins when “we take the robot out into more open environments, such as outdoors, in the home, or on the street,” where the robot must understand its surroundings and operate safely. “Right now, robots are not safe,” he says. “There is no solution to that today.”
Healthcare offers a more personal example for LeBrun. LeBrun’s previous company was Nabla, an AI healthcare startup. He likened today’s AI systems to doctors who only read textbooks and have no training. While LLMs may be useful in medicine, they only cover “only 1% of medicine,” he said. The rest depends on your actual experience.
But LeBlanc said global models cannot be built in a lab. According to the CEO, AMI requires a real environment and close partners to train based on reality. “We need access to the real world,” and “it’s easier when you have a partner.” That’s part of what draws him to Asia, where robots, chips and factories actually exist.
LeBlanc is yet to reveal a complete Asia strategy. “It’s too early,” he said. But the gravitational pull toward South Korea comes down to two things. First, South Korea has advanced industries in robotics, semiconductors, and manufacturing. It’s a hardware-intensive sector that was barely touched by the first wave of AI.
The second attraction is speed. LeBlanc pointed to South Korea’s national plan to invest in AI and its track record as an early adopter. “Twenty-five years ago, South Korea was the first country to adopt the Internet,” he said. He calls this combination — a deep industrial base and a willingness to deploy AI quickly — “unique” and why “we want to be here from day one.”
“I’m telling Alex and the team to come to Korea,” JP Lee, CEO of SBVA and one of AMI’s supporters in Asia, told TechCrunch.
Lee said the government has done a “great job” in funding local sovereign LLM models, which are already “sufficient” for general-purpose tasks, but called on South Korea to continue investing in physical AI as well. “These should coexist,” he said, citing plans to mobilize about $880 billion for chips, AI data centers and physical AI as one of the three pillars announced by the South Korean government in June.
Lee argued that South Korea’s value to foreign companies goes beyond hardware. Local developers quickly adopt and adapt new tools. This pattern has given rise to homegrown internet players such as Naver and Kakao.
Despite having star power and a $1 billion check, AMI still has nothing to sell. The startup, co-founded by Turing Award winner Yann LeCun after leaving Meta, raised $1.03 billion in March at a pre-money valuation of $3.5 billion. There’s no product yet and no timeline he’s committed to. “I’ll give you a surprise when you’re ready,” LeBlanc said.
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
