Yann LeCun, founder of Advanced Machine Intelligence Labs, said his company is developing a new artificial intelligence system called Joint Embedding Predictive Architecture to create more flexible AI than current large language models.

LeCun, who worked at Meta from 2013 to 2025 before establishing AMI Labs in Paris, argues that existing large language models excel at defined tasks such as coding and text generation but lack understanding of real‑world physics. He said the current approach of scaling up these models is unlikely to achieve human‑level or animal‑level intelligence.

The JEPA system builds abstractions of the world that allow an AI to evaluate the outcomes of actions, filtering out irrelevant information. In a demonstration, the model could determine that a pen balanced on its tip would fall in any direction rather than predict a single outcome, according to LeCun. He said the architecture is intended for industrial applications first and could eventually support general purpose AI with minimal fine‑tuning.

AMI Labs announced in early 2024 that it had raised more than $1 billion in a seed round led by investors including Nvidia and a fund managed by Jeff Bezos. The funding, one of the largest seed rounds in Europe, indicates strong investor confidence in the technology.

Ingmar Posner, professor at Oxford University and director of its Applied AI Lab, said the next decade will focus on systems that can explain cause and effect and simulate alternative actions. He and his team have been working on world models for four years, inspired by a 2018 paper by David Ha and Jürgen Schmidhuber. Posner described the goal as creating a mechanistic world model that can organise knowledge for efficient use.

Other groups, including DeepMind with its Genie model, Wayve with Gaia, and World Labs founded by Fei‑Fei Li, are also advancing world‑model research. LeCun noted that AMI Labs will refine its model during the remainder of 2025 and aims to deploy it in industrial settings in 2026.

LeCun emphasized that while autonomous robots will require human input to formulate questions and design tasks, future AI systems will serve as assistants to leaders, similar to how senior staff support executives. The timeline for broader deployment remains uncertain.