Luka Golob and Abdul Sittar

Abstract
Advancements in artificial intelligence and increased internet
accessibility have made it simpler to create and disseminate fake
news with customized content. However, they also improved the
ability to analyze and identify such misinformation. To effectively
train high-performance models, we require high-quality, up-to-
date training datasets. This article delves into the potential for
generating fake news through factual modifications of articles.
This is facilitated by prompt-based content generated by large
language models (LLMs), which can identify and manipulate
facts. We intend to outline our methodology, highlighting both
the capabilities and limitations of this approach. Additionally,
this effort has resulted in new quality synthetic data that can be
incorporated into the standard FAK-ES dataset.