LATIDIA · Investigación

Un giro equivocado no arruina el viaje: autoevolución de habilidades guiadas por la desviación para agentes de LLM

arXiv: 2609.29154v1Tipo de anuncio: nuevo Resumen: Los agentes de modelos de lenguaje grandes dependen cada vez más de las habilidades del lenguaje natural para resolver tareas complejas de uso de herramientas. Sin embargo, tales tareas a menudo admiten múltiples rutas de solución válidas,

La noticia

arXiv:2609.29154v1 Announce Type: new Abstract: Large language model agents increasingly rely on natural-language skills to solve complex tool-use tasks. However, such tasks often admit multiple valid solution paths, making it inappropriate to improve skills by forcing failed trajectories to match a fixed successful trajectory. Moreover, failed trajectories are rarely entirely wrong: an agent may first collect useful evidence and make meaningful progress, but later deviate into an erroneous suffix. We therefore argue that skill self-evolution should identify where productive problem solving begins to break down, rather than reflect coarsely over the entire failure. Based on this insight, we propose SkillPivot, a deviation-point-guided framework for skill self-evolution. SkillPivot detects the transition