Alexandre LeBrun, the chief executive of AMI Labs, told TechCrunch during a stop at the International Conference on Machine Learning in Seoul that his company will not label its work as artificial general intelligence (AGI) or superintelligence. “We never used the word AGI,” LeBrun said, adding that the industry has already moved on to the term superintelligence, which he also finds vague and unhelpful.
Instead of chasing terminology, LeBrun is focused on building a "world model"—an AI system that predicts the next state of the physical world rather than just the next word. He explained that a large language model (LLM) forecasts text, while a world model anticipates how objects will move, such as a glass tipping over when nudged. The two approaches, he said, are complementary: LLMs excel at language processing, whereas world models provide the context needed for robots to understand and act safely in real environments.
LeBrun highlighted the current limits of robotics, noting that most industrial robots follow static routines and lack situational awareness. He gave a hypothetical example of a robot dancing at a public event and inadvertently kicking a child, underscoring the need for context‑aware AI. "Robots are not safe right now," he warned, "there’s no solution for that today." The challenge, he said, intensifies when robots move beyond controlled factory floors into homes or streets.
Healthcare, another sector where LeBrun sees potential, illustrates the gap between language‑only AI and real‑world expertise. He likened today’s AI to a doctor who has only read textbooks and never completed a residency, suggesting that world‑model AI could bridge the missing 99 percent of practical experience.
To train its models, AMI Labs needs access to real‑world environments, a need that is driving the company toward Asia, and South Korea in particular. LeBrun cited Korea’s advanced robotics, semiconductor and manufacturing base, as well as the country’s rapid adoption of new technologies, as key attractions. "Korea was the fastest adopter of the internet 25 years ago," he said, adding that the nation’s national AI plan and substantial funding for chips, data centers and physical AI create a unique ecosystem for AMI Labs.
Local backer JP Lee, CEO of SBVA, echoed the sentiment, praising Korea’s government for supporting sovereign LLMs and urging continued investment in physical AI. Lee pointed to Seoul’s June initiative to mobilize roughly $880 billion for AI‑related infrastructure as evidence of the country’s commitment.
AMI Labs, co‑founded by Turing Award winner Yann LeCun after his departure from Meta, remains pre‑product. The startup raised $1.03 billion in March at a $3.5 billion pre‑money valuation, but it has yet to announce a product timeline. LeBrun promised a surprise when the company is ready, emphasizing that the next phase will involve real‑world testing with industrial partners rather than laboratory demos.
Questo articolo è stato scritto con l'assistenza dell'IA.
News Factory APP - notizie agentiche per potenziare il tuo SEO e AEO.