PANORAMA ACTUAL DE LA SEGURIDAD EN MODELOS DE LENGUAJE EXTENSOS: UN ANÁLISIS INTEGRAL SOBRE ALUCINACIONES Y MECANISMOS DE CONTENCIÓN
FEGLININ · 31.05.2026 · 94-102
DOI: 10.70417/v29cg12eojiv ↗Kurz gefasst
Fachartikel in der Zeitschrift FEGLININ, online erschienen am 31.05.2026. Verfasst von Ing. Marco Tulio Ramirez Reyes und 3 weiteren Autor:innen.
Abstract (Originalsprache)
The recent surge in the use of Extended Language Models (LLMs) in critical production environments reveals inherent weaknesses stemming from their probabilistic nature. Hallucinations and susceptibility to computational attack vectors emerge as symptomatic manifestations of a flawed approach to their training. Through a systematic descriptive-analytical review of specialized literature and metadata in technical ecosystems, contemporary containment architectures, known as "Guardrails," were compared. This review demonstrates that the etiology of hallucinations is rooted in the binary tokenization of pre-training, while threat neutralization requires layered protection schemes that span the training, inference, and emission phases. The study concludes that guardrails deployed as standalone library components are insufficient; their technical potential is realized only through their systemic integration within the software lifecycle under rigorous algorithmic governance frameworks, ensuring organizational risk tolerance.
Zitieren
APA
Reyes, I. M. T. R., Valencia, D. R. E. C., Villegas, D. A. C., & Morales, D. A. F. (2026). PANORAMA ACTUAL DE LA SEGURIDAD EN MODELOS DE LENGUAJE EXTENSOS: UN ANÁLISIS INTEGRAL SOBRE ALUCINACIONES Y MECANISMOS DE CONTENCIÓN. FEGLININ, 94-102. https://doi.org/10.70417/v29cg12eojiv
Vancouver
Reyes IMTR, Valencia DREC, Villegas DAC, Morales DAF. PANORAMA ACTUAL DE LA SEGURIDAD EN MODELOS DE LENGUAJE EXTENSOS: UN ANÁLISIS INTEGRAL SOBRE ALUCINACIONES Y MECANISMOS DE CONTENCIÓN. FEGLININ. 2026:94-102. doi:10.70417/v29cg12eojiv
BibTeX
@article{Reyes2026,
author = {Reyes, Ing. Marco Tulio Ramirez and Valencia, Dr. René Edmundo Cuevas and Villegas, Dr. Arnulfo Catalán and Morales, Dr. Angelino Feliciano},
title = {PANORAMA ACTUAL DE LA SEGURIDAD EN MODELOS DE LENGUAJE EXTENSOS: UN ANÁLISIS INTEGRAL SOBRE ALUCINACIONES Y MECANISMOS DE CONTENCIÓN},
journal = {FEGLININ},
year = {2026},
pages = {94--102},
doi = {10.70417/v29cg12eojiv}
}
EUVIMED