AMI Labs CEO rejects 'AGI' and 'superintelligence' labels in favor of grounded progress

Alexandre LeBrun, CEO of AMI Labs, is taking a refreshingly skeptical stance on the AI industry's obsession with grand terminology. While competitors race to claim breakthroughs in "AGI" and "superintelligence," LeBrun dismisses these labels as hollow marketing.

"There's no good definition. What is superintelligence? I don't know. It's not a very useful word," he told TechCrunch during a recent visit to Seoul. The company founded by Turing Award winner Yann LeCun simply avoids such terminology altogether, preferring to focus on tangible capabilities.

World Models vs. Language Models#

AMI Labs is building world models — systems that predict the next state of the physical world rather than the next word in a sentence. The distinction matters: a language model guesses text sequentially, while a world model understands how reality unfolds. LeBrun likens it to everyday intuition — nudge a glass off a table, and you know it will fall and spill.

Rather than viewing world models and large language models as competitors, LeBrun frames them as complementary tools. Just as human brains compartmentalize language and spatial reasoning, both approaches are needed for AI systems to operate safely in the real world. LLMs excel at processing text, but they falter in physical environments where context and spatial awareness are critical.

The robotics challenge#

Today's robots are largely static — running the same routines repetitively. LeBrun points to a public incident where a dancing robot approached and kicked a child as evidence of AI's fundamental weakness in understanding context. "The hardware is very advanced; progress in hardware in the last few months is incredible, but there's no brain," he said.

Making robots merely "aware of the context" would represent "a very big difference for the world." The real hurdle emerges when machines venture beyond controlled lab settings into homes, factories, or streets. At that point, they must grasp their surroundings and operate safely — something that isn't reliably possible today.

A healthcare parallel#

LeBrun draws on his prior venture, Nabla (an AI health startup), to illustrate the gap. Current AI systems in medicine are like doctors trained only from textbooks without hands-on residency experience. LLMs might handle 1% of healthcare tasks effectively, but the rest depends on real-world experience and judgment that only a world model trained on actual environments can provide.