Code as Worlds: Agentic Discovery of Executable World Representations

_akhaliq · x · 2026-08-31

The paper "Code as Worlds" introduces code as executable representations for the physical world and an agentic process to discover them through iterative simulation and verification. This approach transforms raw observations into reusable physical data—explicit states, dynamics, and mechanisms—capturing not only what was seen but the underlying world that produced it. It provides scalable physical supervision, enabling state-of-the-art performance on quantitative physical reasoning benchmarks.

Related event: Code-as-World: Representing Physical Worlds as Executable Code for Reasoning(2 posts)→

Original post →

More from Research

Research channel →