LATIDIA · Ciberseguridad
Evidencia de marcas de agua en la modificación de LLM de peso abierto: límites exactos para marcas de agua de salida de rango bajo
arXiv:2611.21702v1 Tipo de anuncio: nuevo Resumen: Los modelos lingüísticos de pesos abiertos pueden seguir generando contenido después de que se modifiquen sus pesos publicados, mientras que un verificador de marcas de agua sigue puntuando los resultados en función de la base original y
WhatsApp ↗Telegram ↗
La noticia
arXiv:2611.21702v1 Announce Type: new Abstract: Open-weight language models can continue generating after their released weights are modified, while a watermark verifier still scores outputs against the original base and marked models. We study how such modifications change the evidence available to an OpenStamp-style verifier. At each shared prefix, we represent the modification as a residual in the next-token logits. We derive an exact finite-vocabulary lower bound on the expected log-likelihood-ratio evidence under a residual-span constraint, along with a sharp span-only deviation bound. We also characterize the first-order evidence shift through a covariance term with an explicit second-order remainder, and determine the residual span needed to reach a target evidence level.