Debugging Code World Models

Babak Rahmani · 2026

Read Debugging Code World Models by Babak Rahmani (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-07672.txt.

Description

Code world models are language models trained to simulate program execution by predicting runtime state after each command, offering internal checking as an option to chain-of-thought thinking. Babak Rahmani studies where these models fail, on local execution and long-horizon state tracking.

Two dominant failure regimes emerge on real-code benchmarks: dense state outputs exhaust token budgets on long executions. Failures concentrate in string-valued state, attributed to subword tokenization rather than program structure. Controlled permutation tests probe long-horizon action further. Code model teams get a failure taxonomy that points at representation and budget, not just thinking depth.

Catalog record

CREATORBabak Rahmani
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE36 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-07672

Browse the full library · About the Tower of Babel