GoKawiilWhile the rest of the AI industry races to label its work as “AGI” or “superintelligence,” Alexandre LeBrun, the CEO of Yann LeCun’s world model startup, AMI Labs, avoids the terms altogether. Lebrun said in an interview with TechCrunch that the company doesn’t use terms like “AGI” or “superintelligence” at all.
“We never used the word AGI. And I just noticed that nobody is using it anymore; they switched to superintelligence,” he said. “Next time we’ll switch to something else.” He isn’t sold on the new label either. “There’s no good definition. What is superintelligence? I don’t know. It’s not a very useful word.”
It’s a pointed stance from a founder sitting at the center of AI’s newest race.
TechCrunch talked to LeBrun while he was in Seoul last week for The International Conference on Machine Learning, where he was scouting for local industrial partners, global companies, and researchers. AMI Labs is still pre-product, but it’s already courting robotics, manufacturing, and electronics players. A world model, which incorporates physics to predict and work with the real world, needs to prove itself outside the lab, LeBrun explained.
One area where world models are expected to have a large impact is robotics. For now, robots are just running fixed routines, “completely static,” and AI remains “really dumb in the physical world,” LeBrun said.
Even when AI can merely make robots “aware of the context” that would mark “a very big difference for the world.” Such context-aware AI would have been useful, for example, in preventing a robot that was dancing and doing kung fu at a public event from approaching and kicking a child. “The hardware is very advanced; progress in hardware in the last few months is incredible, but there’s no brain.”
A large language model (LLM) predicts the next word or text, and a world model predicts the next state. Nudge a glass off the table, and you already know it will tip and spill; that’s the intuition a world model is meant to capture: predicting the next state of the world, LeBrun explained.
He isn’t claiming world models are better than LLMs, which are “complementary, not replaceable” when it comes to AI systems that understand the physical world, LeBrun said. Drawing a parallel to the human brain’s distinct language and reasoning functions, he added that LLMs will remain the most efficient tools for processing language while world models will provide context and real-world understanding.
Almost every industry that “touches the real world” could eventually make use of robotics based on world models, LeBrun said, arguing that physical environments remain where LLMs are weakest.
A factory robot repeating the same motion works well enough today, he said. The challenge begins when “you take your robot outside into a more open environment, in your household, or in the street,” where it must understand its surroundings and operate safely. “Robots are not safe right now,” he said. “There’s no solution for that today.”
... continue reading