OSF repository with details, data and code: https://osf.io/jn349/

The present study investigates the impact of tone and impersonation on the production of synthetic disinformation using OpenAI’s Large Language Models (LLMs) davinci-002, davinci-003, gpt-3.5-turbo, and gpt-4.

The primary objective of this study is to investigate the impact of impersonation and tone on the production of synthetic disinformation using LLMs.

The study aims to generate text in the shape of social media posts on various topics by a synthetic disinformation actor named ‘Sam’ and examine the effectiveness of different tones (polite, neutral, impolite) in generating disinformation.

By testing four different LLMs, the study also aims to compare the performance of different LLMs in generating synthetic disinformation.

Ultimately, the study seeks to provide insights into the role of tone and impersonation in generating synthetic disinformation and its implications for the prevention and detection of disinformation campaigns.