While the AI industry races to label "AGI" and "superintelligence," AMI Labs walks a different road. CEO Alexandre LeBrun, who co-founded the startup with Turing Award winner Yann LeCun, skips the buzzwords and tackles what he calls the missing link in modern AI: physical intuition.

Moving Past the Semantic Trap of AGI

At the International Conference on Machine Learning in Seoul, LeBrun said AMI Labs refuses to use terms like "AGI" or "superintelligence." "There’s no good definition. What is superintelligence? I don’t know. It’s not a very useful word," he said, adding that the industry merely swaps labels to keep hype alive.

Instead of chasing theoretical consciousness, AMI Labs builds "world models"—systems that learn the laws of physics and predict the next state of the world. A large language model guesses the next word; a world model knows that nudging a glass off a table makes it tip and spill.

The Missing "Brain" in Robotics

LeBrun points out that hardware in robotics has surged, but software still acts "really dumb in the physical world." Most robots follow static routines and stumble in homes or busy streets.

He treats LLMs and world models as partners, not rivals. Just as the human brain separates language from reasoning, LLMs excel at text while world models supply the context needed for physical interaction. That context matters for safety—for example, it can stop a robot from hurting a person during a complex move.

Scaling Through Industrial Partnerships

AMI Labs is still pre-product, having raised $1.03 billion at a $3.5 billion pre-money valuation. Because world models need real-world data, the company is hunting industrial partners that can open up messy, complex environments.

LeBrun says South Korea makes sense as a hub: its semiconductor, robotics, and manufacturing sectors are world-class. By plugging into "hardware-heavy" industries, AMI Labs can train its models on the chaotic reality that LLMs alone miss.

Why This Matters

The shift from generative text to physical intelligence marks the next AI frontier. The first wave digitized knowledge; the next wave, powered by world models, will digitize physical interaction. Developers and founders should watch for value not in poetry generation but in safe, intuitive navigation of factories and living rooms.

Key Takeaways

  • World Models vs. LLMs: LLMs predict the next token; world models predict the next physical state, giving the intuition needed for real tasks.
  • Physical Context is Critical: Today’s AI lacks the situational awareness for open environments; world models aim to close that gap.
  • Strategic Industrial Focus: AMI Labs targets hardware-centric regions like South Korea to gather the data and partnerships required to train physical AI.