AMI Labs bets on reasoning beyond language models.
PARIS, FRANCE — July 2026. Yann LeCun, the former chief artificial intelligence scientist at Meta, has intensified his challenge to the dominant chatbot industry by presenting AMI Labs as an alternative path toward more capable machine intelligence. The Turing Award-winning researcher argues that systems such as ChatGPT, Claude and Gemini can produce fluent answers without developing a genuine understanding of the physical world. In his assessment, today’s large language models are highly effective at recognizing and reproducing statistical patterns, but they remain unable to reason with the flexibility, memory and causal awareness displayed by humans and animals. His new company is therefore concentrating on systems designed to understand reality, anticipate consequences and plan actions before producing a response.
LeCun has repeatedly rejected the idea that continuously expanding language models will automatically lead to human-level intelligence. He maintains that chatbots are impressive in structured tasks such as programming, language processing and certain mathematical problems because those activities involve patterns that can be represented through enormous quantities of digital information. Their limitations become more visible when they must interpret unpredictable environments, understand physical relationships or determine how an action could transform a real situation. LeCun said these systems can appear intelligent because they manipulate language persuasively, but he argues that linguistic fluency should not be confused with deep comprehension.
AMI Labs is developing what the company calls world models, artificial intelligence systems intended to construct abstract internal representations of environments rather than predict only the next word, image element or data point. The research is closely associated with JEPA, or Joint Embedding Predictive Architecture, an approach that learns relationships between portions of information in a compressed representation space while ignoring details that are irrelevant or impossible to predict. In practical terms, the system would attempt to create an internal simulation of a situation, evaluate several possible developments and select actions according to their expected consequences. This architecture could eventually allow machines to demonstrate persistent memory, common-sense reasoning, planning and greater control when operating in complex physical or industrial settings.
The company raised $1.03 billion in initial financing at a pre-investment valuation of approximately $3.5 billion, placing it among the most heavily funded artificial intelligence startups launched in Europe. Its financing round was supported by major venture capital groups and Bezos Expeditions, the investment organization associated with Amazon founder Jeff Bezos. AMI Labs operates across Paris, New York, Montreal and Singapore, bringing together researchers and engineers with experience at Meta, Google DeepMind and other technology organizations. The laboratory plans to publish research openly and release part of its work through open-source channels while collaborating with academic institutions and industrial partners.
AMI Labs is initially targeting applications in manufacturing, automotive systems, aerospace, healthcare, pharmaceuticals, automation, wearable technology and robotics. These sectors require artificial intelligence to interpret continuous, noisy information generated through cameras, sensors and machinery, where errors can have immediate physical or operational consequences. LeCun has also suggested that world models could eventually support domestic robots capable of navigating homes, understanding ordinary objects and responding safely to unexpected situations. The project does not yet represent a direct consumer chatbot competing feature by feature with ChatGPT, but it constitutes a competing scientific blueprint for how the next generation of artificial intelligence could be constructed.
The next AI revolution may begin by understanding the world, not merely describing it.