AMI Labs CEO Alexandre LeBrun rejects 'AGI' hype, bets on world models

Alexandre LeBrun, CEO of AMI Labs (co-founded with Yann LeCun), refuses to use industry buzzwords like AGI or superintelligence. In a TechCrunch interview, he noted that the AI field has already moved from AGI to superintelligence and may switch again, but neither term has a clear definition. Instead, AMI Labs builds world models—systems that predict the next state of the physical world, much like humans intuit that a glass nudged off a table will fall and break. LeBrun contrasts this with large language models (LLMs), which predict the next word. He sees LLMs and world models as complementary: LLMs handle language efficiently, while world models provide real-world context and understanding.

LeBrun argues that the current AI is “really dumb in the physical world,” especially in robotics where hardware has advanced but robots lack a “brain” to understand their surroundings. He points to incidents like a robot kicking a child during a public demo as evidence that safe, context-aware AI is still missing. To train world models, AMI Labs needs real-world environments—factories, households, streets—which is driving them to partner with industrial players in Asia. LeBrun is actively scouting in South Korea for robotics, semiconductor, and manufacturing partners, attracted by Korea’s advanced hardware industries and fast AI adoption. He calls the combination “unique” and wants to be there from day one. The company is pre-product but aims to bring world models to sectors like healthcare, where LLMs cover only 1% of the needs, and robotics, where adaptive intelligence could prevent accidents.