Over recent years, the spotlight has been on large language models (LLMs), but as we reach their limits, researchers are turning to a new frontier: world models. These systems aim to simulate the physical world, offering potential applications in robotics and beyond.

The journey from chatbots to complex simulations is not straightforward. Experts like Vincent Sitzmann, Anastasis Germanidis, and Ben Mildenhall are leading the charge, but it’s unclear how these systems will interface with humans. Will we interact through chat or directly with a simulated world?

While LLMs have dominated headlines, some in the AI community view world models as their potential successors. However, Yann LeCun, former chief AI scientist at Meta, argues that extending current LLM capabilities to human-level intelligence is unrealistic.

The path forward, according to many, involves tackling specific use cases first—like robotics and research—and then piecing together a broader interface. It’s an ambitious shift, but one that could redefine what we expect from AI systems in the future.