GumbelSoft: Diversified Language Model Watermarking via the GumbelMax-trick

arXiv:2402.12948v1 Announce Type: new
Abstract: Large language models (LLMs) excellently generate human-like text, but also raise concerns about misuse in fake news and academic dishonesty. Decoding-based watermark, particularly the GumbelMax-trick-based watermark(GM watermark), is a standout solution for safeguarding machine-generated texts due to its notable detectability. However, GM watermark encounters a major challenge with generation diversity, always yielding identical outputs for the same prompt, negatively impacting generation diversity and user experience. To overcome this limitation, we propose a new type of …

abstract academic arxiv challenge concerns cs.cl decoding fake fake news generate generated human human-like language language model language models large language large language models llms machine major misuse raise solution text trick type via watermark watermarking