New Grammar-Driven Watermark Improves LLM Code Quality and Detectability

Researchers have developed a new method for watermarking code generated by large language models (LLMs) called Grammar-Driven Watermark (GDW). This approach aims to improve the balance between the quality of the generated code and the detectability of the watermark, which is a challenge with existing methods due to code's low-entropy nature. GDW uses a grammar-guided masking mechanism and assigns different biases to syntax-critical versus content-bearing tokens, enhancing detectability and robustness against attacks like variable renaming. AI

IMPACT This research could lead to more reliable methods for identifying AI-generated code, potentially impacting intellectual property and security in software development.

RANK_REASON The cluster contains an academic paper detailing a new method for code watermarking in LLMs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• Grammar-Driven Watermark
• large-language models
• variable-renaming attacks

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →