What makes a neural operating system?
Actions, internal state, generated observations, and the NeuralOS implementation.
Papers · explanations · citations
Neural operating systems
and GUI world models
Can a neural model learn the behavior and state of an interactive computer?
A reading guide to learned computer simulators, generative interfaces, and interactive world models. Explore the methods, experiments, and connections between papers.
Three ways into the literature
Read in sequence, or follow your question.
From neural game engines to computers
How controls, memory, and visual prediction change when the environment is a desktop.
What should a GUI world model get right?
Visual fidelity, action correctness, persistent state, and useful environments for agents.
A starting point
NeuralOS
First preprint · July 11, 2025
Revised · March 12, 2026
ICLR 2026
Towards Simulating Operating Systems via Neural Generative Models
NeuralOS studies learning an interactive computer environment from user-input and screen sequences. Its recurrent state model and diffusion renderer simulate operating-system interfaces without executing the underlying applications during simulation.
The original paper develops the modeling setting, state tracking, and precise interaction. The revised paper also demonstrates learning an application from synthetic demonstrations, pointing toward interfaces learned from examples.
A short reading path
NeuralOS → GameNGen → DIAMOND → ViMo → Code2World → OSWorld
Follow the conceptual path above, or browse the full list in chronological order. Each entry describes the model’s predictions, actions, state, and relationship to NeuralOS.