CodeVision: Detecting LLM-Generated Code Using 2D Token Probability Maps and Vision Models

Zhenyu Xu, Victor S. Sheng · 2025

Read CodeVision: Detecting LLM-Generated Code Using 2D Token Probability Maps and Vision Models by Zhenyu Xu, Victor S. Sheng (2025) 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-2501-03288.txt.

Description

Zhenyu Xu and Victor S Sheng present a 2025 arXiv paper on telling machine-written code from human code. Their idea turns each token's probability into a two-dimensional map, an image that keeps the code's layout, indenting, and brackets visible. Vision models such as transformers and ResNet then read the map for telltale patterns.

The method held up across several programming languages and beat traditional detectors in their tests. Because it works from probabilities rather than watermarks or retrained language models, it stays light to compute. The authors aim it at academic integrity settings, where flagged code mixes content and structure clues.

Catalog record

CREATORZhenyu Xu, Victor S. Sheng
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE32 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-03288

Browse the full library · About the Tower of Babel