Freedom to Play Against the Apparatus: Hallucinations in Synthetic Text
Bakhtiniana: Revista de Estudos do Discurso · 2026 · 21 (3)
DOI: 10.1590/2176-4573e70902 ↗Kurz gefasst
Fachartikel in der Zeitschrift Bakhtiniana: Revista de Estudos do Discurso, 2026. Verfasst von Pollyanna Ferrari und 1 weiteren Autor:innen. Die Arbeit nennt 26 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
The textual production of GPT-type models is characterized by the simulation of realism and by ideological restrictions, both of which are applied to the model through the usage of reinforcement learning from human feedback (RLHF). These filters seem to reinforce an anthropocentric and normative project of control over language. By exploring the phenomenon of jailbreaking - capable of circumventing these filters - this article discusses the extent to which this practice is cabable of “throwing light in the black-box,” based on Arlindo Machado and Vilém Flusser’s critics on anthropocentrism. Finally, some case studies of the application of the jailbreaking procedure are presented as methodology. The extent to which these texts are able to criticize the filters imposed in GPT models is debated.
Zitieren
APA
Ferrari, P., & Bucker, G. B. (2026). Freedom to Play Against the Apparatus: Hallucinations in Synthetic Text. Bakhtiniana: Revista de Estudos do Discurso, 21(3). https://doi.org/10.1590/2176-4573e70902
Vancouver
Ferrari P, Bucker GB. Freedom to Play Against the Apparatus: Hallucinations in Synthetic Text. Bakhtiniana: Revista de Estudos do Discurso. 2026;21(3). doi:10.1590/2176-4573e70902
BibTeX
@article{Ferrari2026,
author = {Ferrari, Pollyanna and Bucker, Gustavo Bonassa},
title = {Freedom to Play Against the Apparatus: Hallucinations in Synthetic Text},
journal = {Bakhtiniana: Revista de Estudos do Discurso},
year = {2026},
volume = {21},
number = {3},
doi = {10.1590/2176-4573e70902}
}
EUVIMED