LATIDIA · Investigación
DACE: Coevolución adversaria impulsada por la diversidad para una alineación robusta de la seguridad de los LLM
arXiv:2611.22278v1 Tipo de anuncio: nuevo Resumen: Garantizar una alineación segura y sólida de los grandes modelos de lenguaje (LLM) es cada vez más difícil a medida que los ataques adversarios evolucionan y superan a los procesos de alineación basados en datos recopilados previamente.
WhatsApp ↗Telegram ↗
La noticia
arXiv:2611.22278v1 Announce Type: new Abstract: Ensuring robust safety alignment of large language models (LLMs) is increasingly difficult as adversarial attacks evolve and outpace alignment pipelines built on pre-collected data. Existing co-evolutionary frameworks let the defender chase a moving attacker, yet their training dynamics expose two compounding pathologies: \emph{attack strategy collapse}, where the attacker overfits to a narrow set of high-reward rewrites, and \emph{defense adversarial forgetting}, where the defender loses competence against earlier attacks as the attack distribution drifts. Existing remedies are partial: attacker-side diversity rewards score textual novelty rather than strategy novelty, and defender-side replay relies on discrete judge scores that ignore estimation uncertainty, response stochasticity, and defender non-stationarity. We