While the AI industry races to brand everything as "superintelligence," one CEO backed by Meta's chief AI scientist is pumping the brakes on the hype. Alexandre LeBrun, who leads AMI Labs - a world model startup co-founded with Yann LeCun - is taking a hard stance against the terminology that's become Silicon Valley's favorite buzzword. In an industry where every startup claims they're building AGI, LeBrun's refusal to use the term signals a philosophical split that could reshape how we talk about AI's future.
AMI Labs is charting a different course in an industry drunk on its own superlatives. While OpenAI counts down to AGI and Google DeepMind promises artificial general intelligence within years, Alexandre LeBrun won't touch the terminology with a ten-foot pole.
The French entrepreneur, who previously sold his AI startup Wit.ai to Meta in 2015, now leads a venture that's got serious credibility behind it. Co-founded with Yann LeCun - Meta's chief AI scientist and Turing Award winner - AMI Labs represents a contrarian bet that the path to advanced AI doesn't run through the AGI marketing playbook.
"We're not chasing labels," LeBrun told TechCrunch in the interview. The dismissal isn't just semantic pickiness. It's a fundamental disagreement about what AI research should prioritize and how companies should position their work.
AMI Labs is building what's called world models - AI systems that can predict and simulate how the physical world works rather than just processing text and generating responses. LeCun has been vocal that this approach, not scaling up large language models, represents the real path to human-level intelligence. But even he and LeBrun won't use the AGI term to describe their endgame.
The terminology wars in AI have reached fever pitch. OpenAI literally has AGI in its charter and has structured its entire corporate governance around the moment it achieves this milestone. Google DeepMind CEO Demis Hassabis regularly discusses AGI timelines in interviews. Microsoft and Nvidia executives throw around "superintelligence" in earnings calls.
But LeBrun's skepticism taps into a growing backlash among AI researchers who think the hype has gotten out of hand. The terms AGI and superintelligence carry so much baggage - sci-fi connotations, existential risk fears, regulatory implications - that some scientists believe they obscure more than they clarify.
"When you say AGI, everyone imagines something different," one AI researcher who requested anonymity told us. "For some it's human-level performance on benchmarks. For others it's consciousness. For VCs it's whatever will get their portfolio companies valued at $10 billion."
AMI Labs' approach with world models does represent a genuine technical departure from the transformer-based large language models that power ChatGPT and similar systems. Instead of predicting the next word in a sequence, world models try to predict the next state of a system - how objects move, how actions have consequences, how the world evolves over time.
LeCun has argued that this kind of predictive learning is how babies and animals learn about the world, and it's fundamentally more efficient than the text-based pretraining that requires scraping the entire internet. If he's right, AMI Labs could leapfrog the LLM giants without needing their massive compute budgets.
But the company faces steep competition. Google DeepMind has its own world model research with projects like Genie. OpenAI is exploring video generation models that implicitly learn physical dynamics. Even Meta's internal AI teams, separate from AMI Labs, are working on similar problems.
What's striking is that LeBrun and LeCun seem more interested in being right about the science than being first to claim an AGI crown. In an industry where Tesla CEO Elon Musk predicts AGI by next year and Nvidia CEO Jensen Huang says it's five years away, that kind of restraint almost seems radical.
The terminology debate also has real stakes beyond marketing. How we define AGI determines when OpenAI's board decides the company has achieved its mission and triggers major governance changes. It influences which research gets funded and which regulatory frameworks governments adopt.
If "superintelligence" becomes the standard term, it implies we're building something that will surpass human capabilities across the board - a framing that invites heavy government oversight. If instead we talk about "specialized AI systems" or "world models," the conversation shifts to more grounded questions about what these tools can actually do.
LeBrun's stance might also be strategic positioning for AMI Labs. By refusing to play the AGI hype game, the company differentiates itself as the serious, science-first alternative to the breathless promises coming from bigger competitors. It's a bet that eventually the market will reward substance over superlatives.
The approach mirrors LeCun's own brand within AI research. While he's won every major award in the field and leads AI at one of the world's most valuable companies, he's spent recent years as the industry's most prominent skeptic of AGI doomerism and LLM-centric approaches.
LeBrun's refusal to embrace AGI terminology isn't just contrarian posturing - it's a signal that not everyone in AI believes the current narrative. As the industry sorts out what these systems can actually do versus what we call them, AMI Labs is making a bet that world models and scientific rigor will matter more than marketing superlatives. Whether that approach can compete with the hype machines at OpenAI and Google remains to be seen, but it's a reminder that the loudest voices in AI aren't always the most credible. For an industry that keeps moving goalposts on what constitutes intelligence, maybe it's refreshing to have at least one CEO who won't play the definition game at all.
