According to LeCun, the system creates abstract representations of the physical world, allowing AI to focus on useful information rather than attempting to predict every possible outcome.
LeCun said this limitation makes current language models unsuitable for robotics.
Research into world models has expanded in recent years.
LeCun's AMI Labs is pursuing JEPA, while other organisations are working on similar approaches, including Google DeepMind's Genie, Wayve's Gaia and World Labs, the company founded by AI researcher Fei-Fei Li.
LeCun said AMI Labs plans to spend the rest of the year refining its AI system before deploying it in industrial settings next year.
Former Meta chief AI scientist and AMI Labs founder Yann LeCun has challenged the idea that simply making large language models bigger will lead to human-like intelligence, arguing that current AI systems are fundamentally limited when it comes to understanding and interacting with the physical world, BBC reported.
Speaking on the sidelines of the VivaTech conference in France, LeCun said AI models such as ChatGPT, Claude and Gemini are effective at tasks like coding, solving mathematical problems and generating text, but lack the ability to reason about real-world environments.
"We don't have robots that are nearly as good at understanding the physical world as a rat," LeCun said, adding that existing large language models are "not a path towards human level or human-like intelligence, or even animal-like intelligence" because they are not designed to work with real-world data.
Also Read: AI pioneer Yann LeCun challenges Amodei’s job loss claims, flags limits of tech expertise
Instead of relying on large language models, Paris-based AMI Labs is building a different AI architecture known as Joint Embedding Predictive Architecture, or JEPA. According to LeCun, the system creates abstract representations of the physical world, allowing AI to focus on useful information rather than attempting to predict every possible outcome.
To explain the difference, LeCun used the example of balancing a pen on its tip. While a person understands the pen will fall without trying to predict the exact direction, he argued that a language model may still attempt to produce a single statistically likely answer instead of recognising that the direction cannot be determined in advance.
LeCun said this limitation makes current language models unsuitable for robotics. "LLMs are largely hopeless for robotics," he said, rejecting claims that simply scaling up today's models will eventually produce superhuman intelligence.
AMI Labs has attracted significant investor backing for its approach. Earlier this year, the company announced a seed funding round of more than $1 billion with support from investors including Nvidia and the investment fund managing Jeff Bezos' private wealth. The funding is among the largest seed rounds raised by a European startup.
The broader robotics industry is also looking for AI systems that can better understand physical environments, as companies continue investing heavily in humanoid robots while facing challenges in training them to perform everyday household tasks safely and reliably.
Oxford University professor Ingmar Posner, who leads its Applied AI Lab and is also an Amazon Scholar, said future AI systems need to move beyond language generation and explain how the world works.
According to Posner, the next generation of AI should answer questions about cause and effect, identify what matters in a situation and predict how different actions would change outcomes. His research team has spent the past four years developing what he calls a mechanistic world model to organise knowledge so it can be recalled, combined and adapted more effectively.
Research into world models has expanded in recent years. LeCun's AMI Labs is pursuing JEPA, while other organisations are working on similar approaches, including Google DeepMind's Genie, Wayve's Gaia and World Labs, the company founded by AI researcher Fei-Fei Li.
LeCun said AMI Labs plans to spend the rest of the year refining its AI system before deploying it in industrial settings next year. If those efforts succeed, he believes the technology could eventually support more general-purpose intelligence across a wide range of applications with limited additional training.
He also said humans would continue to play the central role in deciding what problems AI should solve.
"Our interaction with future AI systems, even if they are smarter than us, is going to be like the interaction between a captain of industry or a political leader with their staff of assistants, many of whom are smarter than they are," LeCun said.
First Published on July 4, 2026, 16:01:02 IST
