From the source
Frequently, and badly.
We need a better taxonomy, rather than calling everything a world model Jeff Hawke February 27th, 2026 The past year has seen enormous interest arise from the research community and different industries.
This is visible in machine learning conferences, new startups, and even in Google search trends.
These trends show signs of acceleration, not slowing.
New papers, demos, and systems are appearing every week.
That momentum is meaningful.
It means a fast-growing number of very smart people, who could spend their time on almost anything, believe world models are where their time is best spent.
So, what is being modeled?
With this lens, I'd like to propose a taxonomy of four categories of AI model currently being described as a ‘world model.’
World models: “ learn how the world evolves ” Spatial intelligence: “ learn how the world appears ” Behaviour models: “ learn how to act within in a world ” Proxy world models: “ learn an abstraction of the world ” The architecture of a world model World Models: “Learn How the World Evolves” …
