LATIDIA · Ciberseguridad
Límites fundamentales de las marcas de agua estadísticas para los LLM con aleatoriedad común fija y optimizable
arXiv:2611.19305v1 Tipo de anuncio: cruzado Resumen: La marcación de agua se ha convertido en una técnica importante para distinguir el texto generado por IA del texto escrito por humanos. Sin embargo, la mayoría de los trabajos existentes son principalmente empíricos y se basan en
WhatsApp ↗Telegram ↗
La noticia
arXiv:2611.19305v1 Announce Type: cross Abstract: Watermarking has emerged as an important technique for distinguishing AI-generated text from human-written text. However, most existing work is primarily empirical, relies on ad hoc watermarking designs, and lacks a fundamental understanding of the performance limits. In this paper, we propose a unified theoretical framework for statistical watermarking of large language models (LLMs), casting the problem as an independence test between generated text and common randomness and jointly designing the watermarking embedder and the detector. Our objective is to maximize detection power while controlling the worst-case false positive rate (FPR) and limiting distortion to text quality. A central focus of our framework is the design