Résumé:
16
Corresponding author email : ouissem.touameur@gmail.com
From Hallucinations to Truth: Strategies for Improving Chatbot
Accuracy and Trustworthiness
Ouissem, Touameur*; Harrag, Fouzi
Farhat Abbas, Setif University
Abstract
Hallucinations in chatbot systems—instances where the model generates
inaccurate or entirely fabricated information—pose a significant challenge to the
reliability and trustworthiness of AI-driven communication. This paper provides
a comprehensive review of the state-of-the-art in tackling hallucinations,
detailing their underlying causes, varied manifestations, and the landscape of
existing mitigation strategies. We critically examine leading approaches,
including knowledge augmentation models, advanced fine-tuning methods,
automated fact-checking systems, and the emerging role of explainable AI in
identifying and reducing hallucinations
