While the rest of the AI industry struggles to label its work as “AGI” or “superintelligence,” Alexandre LeBrun, CEO of global model startup Yann LeCun, AMI Labs, avoids terms altogether. Lebrun told TechCrunch that the company doesn’t use terms like “AGI” or “superintelligence” at all.
“We never used the word AGI. And I just noticed that nobody uses it anymore; they changed to superintelligence,” he said. “Next time we’ll switch to something else.” Not sold on the new label either. “There is no good definition. What is superintelligence? I don’t know. It’s not a very useful word.”
It’s a sharp stance from a founder who sits at the center of AI’s newest battle.
TechCrunch spoke with LeBrun while he was in Seoul last week for the International Conference on Machine Learning, where he was looking for local industry partners, global companies and researchers. AMI Labs is still pre-product, but it’s already courting robotics, manufacturing, and electronics gamers. A global model, which incorporates physics to predict and work with the real world, needs to be proven outside the lab, LeBrun explained.
One area where global models are expected to have a big impact is robotics. For now, robots are just performing fixed routines, “totally static,” and AI remains “really dumb in the physical world,” LeBrun said.
Even when AI can just make robots “context aware” that would make “a very big difference to the world”. Such context-aware artificial intelligence would be useful, for example, in prevention a robot that danced and did kung fu at a public event from approaching and kicking a child. “The hardware is very advanced; the progress in hardware in the last few months is incredible, but there is no mind.”
A large language model (LLM) predicts the next word or text and a global model predicts the next state. Push a glass off the table and you already know it’s going to tip over and spill. This is the intuition that a global model is meant to capture: to predict the next state of the world, LeBrun explained.
He doesn’t claim that global models are better than LLMs, which are “complementary, not replaceable” when it comes to artificial intelligence systems that understand the physical world, LeBrun said. Drawing a parallel with the separate language and logic functions of the human brain, he added that LLMs will remain the most effective tools for language processing, while global models will provide the context and understanding of the real world.
Almost any industry that “touches the real world” could eventually make use of world model-based robotics, LeBrun said, arguing that physical environments remain where LLMs are weakest.
A factory robot that repeats the same motion works quite well today, he said. The challenge begins when you “take your robot out into a more open environment, in your home or on the street,” where it needs to understand its environment and operate safely. “Robots are not safe right now,” he said. “There is no solution for this today.”
Health care offers a more personal example for LeBrun, whose previous company was Nabla, a health startup. He likened today’s AI systems to a textbook-trained, unlicensed doctor. LLMs can be useful in medicine, he said, but cover “only 1% of health care.” The rest depends on real world experience.
But a global model, LeBrun said, can’t be built in a lab. To actually train, AMI needs real environments and close partners, according to the CEO. “We need access to the real world,” and it’s “easier for us to do that with partners.” That’s part of what draws him to Asia, where the robots, chips and factories actually are.
LeBrun is yet to define a full strategy for Asia. “It’s too early,” he said. But the attraction to South Korea comes down to two things. First, Korea has advanced industries in robotics, semiconductors, and manufacturing. the hardware-heavy sectors just touched by the first wave of artificial intelligence.
The second attraction is speed. LeBrun pointed to Korea’s national plan to put money into artificial intelligence and its track record as an early adopter. “Korea was the fastest adopter of the Internet 25 years ago,” he said. It’s that combination, a deep industrial base and a willingness to embrace AI quickly, that he calls “unique” and the reason “we want to be here from day one.”
“I was telling Alex and the team to come to Korea,” JP Lee, CEO of SBVA and one of AMI’s supporters in Asia, told TechCrunch.
The government has done “a tremendous job” of funding locally-dominant LLM models, Lee said, and these are already working “pretty well” for general-purpose work, but he is pushing for Korea to continue investing in physical AI as well. He points to Seoul’s plan in June to mobilize about $880 billion for chips, AI data centers and physical AI as one of its three stated pillars: “They should co-exist.”
Korea’s value to foreign companies, Lee argued, is not just in hardware. Local developers are quick to adopt and adapt new tools, a pattern that has spawned homegrown Internet players like Naver and Kakao.
Despite the star power and billion dollar check, AMI has nothing to sell yet. The startup, which was founded by Turing Award winner Yann LeCun after he left Meta, raised $1.03 billion in March at a preliminary valuation of $3.5 billion. No product yet, no timeline to commit to. “We’ll have a surprise when we’re ready,” LeBrun said.
When you purchase through links in our articles, we may earn a small commission. This does not affect our editorial independence.
