The loudest word in AI right now is “superintelligence.” Alexandre LeBrun, CEO of Yann LeCun’s world model startup AMI Labs, won’t touch it. In an interview with TechCrunch AI, LeBrun said his company skips the marketing vocabulary entirely: “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’s not impressed with the replacement either. “There’s no good definition. What is superintelligence? I don’t know. It’s not a very useful word.”
That’s a striking position from a founder who just raised $1.03 billion at a $3.5 billion pre-money valuation with no product to show. What stands out here is that LeBrun is betting the vocabulary vacuum works in his favor. While everyone else sells a destination, he’s selling a specific technical gap.
The gap he’s pointing at
LLMs predict the next word. World models predict the next state. Nudge a glass off a table and you already know what happens next. That intuition is what AMI is trying to build, according to TechCrunch AI’s reporting.
LeBrun’s diagnosis of robotics is blunt. Robots today run fixed routines and are “completely static.” AI is “really dumb in the physical world.” His example is uncomfortable and precise: a robot doing kung fu at a public event that approached and kicked a child. Context awareness alone, he says, would mark “a very big difference for the world.”
The line that should stick with anyone building physical AI: “The hardware is very advanced; progress in hardware in the last few months is incredible, but there’s no brain.”
Complementary, not a replacement
Here’s where LeBrun separates himself from the usual startup pitch. He isn’t claiming world models beat LLMs. He calls them “complementary, not replaceable,” drawing a parallel to how the human brain splits language and reasoning. LLMs stay the best tool for language. World models supply physical context.
His healthcare analogy lands hard, and he’d know, since his previous company was AI health startup Nabla. Today’s AI is a doctor trained only on textbooks with no residency. LLMs cover “only 1% of healthcare.” The rest is real-world experience.
Why Asia, and why now
AMI is pre-product but already courting robotics, manufacturing, and electronics partners. LeBrun was in Seoul for ICML scouting industrial partners. The logic is simple: “We need access to the real world,” and it’s “easier for us to do that with partners.”
Korea gets the attention for two reasons:
- Industrial depth. Robotics, semiconductors, and manufacturing. The hardware-heavy sectors the first AI wave barely touched.
- Adoption speed. “Korea was the fastest adopter of the internet 25 years ago,” LeBrun said.
JP Lee, CEO of SBVA and an AMI backer, told TechCrunch AI he’s been pushing LeBrun toward Korea. Lee credits the government with a “tremendous job” funding sovereign LLMs that work “well enough” for general tasks, and points to Seoul’s June plan to mobilize roughly $880 billion for chips, AI data centers, and physical AI. His view on LLMs and physical AI: “They should coexist.”
What this signals for the next two years
The naming fight isn’t cosmetic. It tracks where the money is moving. When labels get vague, valuations attach to narrative. When labels get specific, they attach to deployment. LeBrun is placing his bet on the second phase arriving.
Expect three things to shift:
- Benchmarks move off the leaderboard and into the warehouse. “Robots are not safe right now. There’s no solution for that today,” LeBrun said. Whoever solves open-environment safety sets the standard.
- Data access becomes the moat. You can’t train a world model in a lab. Companies with factories, fleets, and physical sites hold leverage they haven’t priced yet.
- Geography matters again. Asia’s industrial base gives it a structural edge in physical AI that no amount of GPU spend replicates.
What to do about it
- If you run operations touching the physical world, audit what environmental data you’re already generating and throwing away. That’s the asset.
- If you’re evaluating AI vendors, ask which problems are language problems and which are state-prediction problems. Different tools.
- Treat the vocabulary as a signal. Founders reaching for the biggest word usually have the least to demo.
AMI still has nothing to sell. No timeline either. “We’ll make a surprise when we’re ready,” LeBrun said. Full details are at the original source.