About · Edition III · October 2026
Why an atlas, and who wrote it.
A field survey of learned world models, in one voice, told three ways.
What this is
A chart of one field.
A world model is a small copy of the world that a machine keeps in its head. It guesses what comes next. It plans against the guess before it acts.
This atlas charts that field. It has six provinces and 21 territories. Each territory is told at three altitudes, so there are 63 tellings in all.
A textbook explains a subject once. An atlas lets you pick your altitude. You can come back to the same territory later from a different one.
How it is written
I write every territory three times.
The facts do not change between tellings. The register does. I pick the reader at the top of the page and the whole atlas follows.
Each territory opens with a living diagram of its idea. A still from the same observatory at first light sits beside it. Six of the hardest ideas also have a short film.
Every territory ends with a quick check. You can turn any territory into mission cards, one idea per card.
The three readers
Pick the one that sounds like you.
-
Sonu, 10
One picture, no big words and a thing to try at home. Sonu is my daughter. She is ten.
-
Senior AI Architect
The mechanism, the trade-off, the failure mode and what to measure. For the person who has to ship it.
-
My CIO
The consequence, the risk, the money and the one decision. For the person who decides and pays.
Source
Where the map came from.
CIS 6280, World Models, informed the curriculum arc. Jiatao Gu teaches it at the University of Pennsylvania. The text is mine.
The author
I am FNU Ratnavarma.
I am an independent applied AI researcher and enterprise AI practitioner with 21 years in enterprise technology. I lead AI solution delivery at a global consultancy. My work centers on large US utilities.
Utilities taught me one thing about world models. You cannot try an action on a live grid to see what happens. You need a model that predicts first. That is why I wrote this atlas.
My research asks how AI agents can plan, predict and act safely in high-stakes operations. I work on world models, joint embedding predictive architectures (JEPA) and reinforcement learning. I also test how reliable autonomous agents really are.
I review papers for journals and workshops. I judge AI hackathons and research sprints. I am an IEEE Senior Member and I mentor through the IEEE Power & Energy Society.
-
MS, Information Technology Management Western Governors University
2019 -
BE, Civil Engineering Visvesvaraya Technological University
2005
Identifiers
Find me elsewhere.
- 0009-0008-2191-6575
- ~FNU_Ratnavarma1
- in/ratnavarma
- REZ-5601-2026
Other sites
More from me.
This atlas is a companion to The Notebook, my public notebook on AI and the systems around it.
Questions or ideas? Write to me