On the Reliability of Watermarks for Large Language Models

arXiv:2306.04634v4 Announce Type: replace-cross
Abstract: As LLMs become commonplace, machine-generated text has the potential to flood the internet with spam, social media bots, and valueless content. Watermarking is a simple and effective strategy for mitigating such harms by enabling the detection and documentation of LLM-generated text. Yet a crucial question remains: How reliable is watermarking in realistic settings in the wild? There, watermarked text may be modified to suit a user's needs, or entirely rewritten to avoid detection. We study …

arxiv bots cs.cl cs.cr cs.lg detection documentation flood generated internet language language models large llm llms machine media question reliability simple social social media spam strategy text watermarking watermarks