/disinfo-costs

Primary LanguageJupyter Notebook

Repository for A Cost Analysis of Generative Language Models and Influence Operations

This repository includes the code, LaTeX files, and figures associated with "A Cost Analysis of Generative Language Models and Influence Operations," which is accessible at https://arxiv.org/abs/2308.03740. All code and analysis is contained within codebook.ipynb. Note that the final versions of Figures 2 and 4 in the paper were produced in codebook.ipynb but then manually annotated; both the raw outputs of the code notebook and the edited image files are present in figures. In addition, the file Tweet_counts.txt provides a list of the file sizes of various state-affiliated campaigns identified and publicized by Twitter Transparency at https://transparency.twitter.com/en/reports/moderation-research.html, as well as approximate estimates for the number of tweets associated with each state.

Special thanks to Moulay Akhloufi for providing template LaTeX files in arXiv style, which can be accessed at https://www.overleaf.com/latex/templates/arxiv-and-prime-ai-style-template/qdnhqytdqzsc. This template was itself adapted from templates created by George Kour and available at https://github.com/kourgeorge/arxiv-style.

For questions, please contact Micah Musser at mrm311 [at] georgetown [dot] edu.