Watermarking Language Models with Error Correcting Codes
This addresses the need for reliable detection of AI-generated content, which is crucial for applications like content moderation and authenticity verification, though it appears incremental as it builds on existing watermarking approaches.
The authors tackled the problem of distinguishing machine-generated text from human text by proposing a watermarking framework that uses error correcting codes, which introduces no noticeable quality degradation and is robust to edits, deletions, and translations.
Recent progress in large language models enables the creation of realistic machine-generated content. Watermarking is a promising approach to distinguish machine-generated text from human text, embedding statistical signals in the output that are ideally undetectable to humans. We propose a watermarking framework that encodes such signals through an error correcting code. Our method, termed robust binary code (RBC) watermark, introduces no noticeable degradation in quality. We evaluate our watermark on base and instruction fine-tuned models and find that our watermark is robust to edits, deletions, and translations. We provide an information-theoretic perspective on watermarking, a powerful statistical test for detection and for generating $p$-values, and theoretical guarantees. Our empirical findings suggest our watermark is fast, powerful, and robust, comparing favorably to the state-of-the-art.