Communication Dans Un Congrès Année : 2024

Bias, Subjectivity and Norm in Large Language Models

Résumé

This article reevaluates the concept of bias in Large Language Models, highlighting the inherent and varying nature of these biases and the complexities involved in post hoc adjustments to meet legal and ethical standards. It argues for shifting the focus from seeking bias-free models to enhancing transparency in filtering processes, tailored to specific use cases, acknowledging that biases reflect societal values.
Fichier principal
Vignette du fichier
aequitas.pdf (564.04 Ko) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-04838836 , version 1 (15-12-2024)

Licence

Identifiants

  • HAL Id : hal-04838836 , version 1

Citer

Thierry Poibeau. Bias, Subjectivity and Norm in Large Language Models. Aequitas (Fairness and Bias in AI), Oct 2024, Saint Jacques de Compostelle, Spain. ⟨hal-04838836⟩
14 Consultations
12 Téléchargements

Partager

More