Authors: Yuling Shi, Hongyu Zhang, Chengcheng Wan, Xiaodong Gu

Published on: January 12, 2024

Impact Score: 8.22

Arxiv code: Arxiv:2401.06461

Summary

• What is new: DetectCodeGPT, which specifically targets structural patterns in code for improved detection of machine-generated source code.
• Why this is important: Existing methods like DetectGPT struggle to differentiate between machine and human-authored code due to lack of focus on code-specific patterns.
• What the research proposes: A new method, DetectCodeGPT, which analyzes code attributes and uses strategic perturbations like inserting spaces and newlines to improve detection accuracy.
• Results: DetectCodeGPT significantly outperforms current state-of-the-art techniques in identifying machine-generated code.

Technical Details

Technological frameworks used:

Models used: DetectGPT and novel DetectCodeGPT method

Data used: Code attributes such as length, lexical diversity, and naturalness

Potential Impact

Code generation platforms, cybersecurity firms, and companies concerned with software integrity and authenticity.

Want to implement this idea in a business?

We have generated a startup concept here: AuthentiCode.