Edit Distance Robust Watermarks for Language Models

arXiv:2406.02633v1 Announce Type: cross
Abstract: Motivated by the problem of detecting AI-generated text, we consider the problem of watermarking the output of language models with provable guarantees. We aim for watermarks which satisfy: (a) undetectability, a cryptographic notion introduced by Christ, Gunn & Zamir (2024) which stipulates that it is computationally hard to distinguish watermarked language model outputs from the model's actual output distribution; and (b) robustness to channels which introduce a constant fraction of adversarial insertions, substitutions, and deletions …

abstract ai-generated text aim arxiv cs.ai cs.cr cs.lg edit generated language language models notion output problem robust text type watermarking watermarks