An essay on the history, theory, progress, and potential of world models, a prominent theme at Nvidia GTC 2026, co-written by General Intuition CEO Pim de Witte
Welcome to the 458 newly Not Boring people who have joined us since our last essay! Join 260,170 smart, curious folks by subscribing here:
The real world is - or was - uncomputable. If language and code, two of mankind's most powerful inventions, are inadequate to represent our real world, what do we have left? The Answer is World Models They offer a way to do non-deterministic compute efficiently, and to run [image…
This is very cool. In the World Models piece today, Pim and I wrote that Pi's VLAs are a pragmatic approach to embodied AI and that the company seems to be making a very strategic bet. They keep unhobbling VLAs. [image]
Adding a few things that didn't make the final cut as well: Game and physics engines remain hugely important - they offer determinism, are designed for entertainment, and are optimized for delivering on mass consumer hardware. World Models are probabilistic and currently designed
Packy and I spent the past month unpacking world models from first principles. This piece is the result of that exploration. We go into why, what, how, and look out into the future on the implications of our work.
This is the best primer on arguably the most important frontier in AI: World Models. LLMs are exquisite manipulators of symbols — but they operate entirely in the realm of representation. They can describe clapping, but they cannot clap. The core insight behind World Models is