A Watermark for Large Language Models

arXiv:2301.10226v4 Announce Type: replace-cross
Abstract: Potential harms of large language models can be mitigated by watermarking model output, i.e., embedding signals into generated text that are invisible to humans but algorithmically detectable from a short span of tokens. We propose a watermarking framework for proprietary language models. The watermark can be embedded with negligible impact on text quality, and can be detected using an efficient open-source algorithm without access to the language model API or parameters. The watermark works by …

abstract arxiv cs.cl cs.cr cs.lg embedded embedding framework generated humans language language models large language large language models proprietary text tokens type watermark watermarking