Skip to content

Latest commit

 

History

History
61 lines (48 loc) · 3.05 KB

README.md

File metadata and controls

61 lines (48 loc) · 3.05 KB
Raw

Fact-checking information from large language models can decrease headline discernment

Paper

For more details, you can find the paper here. It should be cited as:

@article{deverna2023artificial,
  title={Fact-checking information generated by a large language model can decrease news discernment},
  author={DeVerna, Matthew R. and Yan, Harry Yaojun and Yang, Kai-Cheng and Menczer, Filippo},
  journal={Proceedings of the National Academy of Sciences (Forthcoming)},
  year={2023}
}

Project aim

We conduct a preregistered experiment to investigate whether fact checks provided by a large language model (ChatGPT) can serve as an effective misinformation intervention.

Directories

  • code: scripts to generate results, figures, etc.
  • data: data used in the project
  • environment: python environment files
  • figures: all generated figures
  • results: output files including processed data and statistical reports
  • prompt_engineering: contains everything for the prompt engineering analysis in the supplementary information (the "Accuracy of different prompt methods" secton within the SI).

Requirements

  • Python: see the environment/ directory.
    • Used for most data wrangling/manipulation/basic stats
  • R: version 4.3.0 (2023-04-21)
    • Used for regression analyses

Replication

We utilize both Python and R coding languages in this project.

Python analysis and figure generation

To replicated the Python analysis, please set up an environment as described in the environment/ directory. Then, you should be able to run the below code (after changing your current directory to wherever this README.md file is saved) to run all analyses and generate all figures:

cd code
bash run_pipeline.sh

R analysis and figure generation

The version of R that is utilized in this project is: R version 4.3.0 (2023-04-21). We also utilized RStudio Version 2023.06.0+421 (2023.06.0+421). All analyses and figures are created with the RMarkdown files in the code/r_code/ directory.

After installing the versions of R and RStudio indicated above, you should be able to open the RMarkdown files with RStudio and "Knit" each one, creating the HTML version currently saved in the same location.

Prompt engineering supplementary analysis

To generate the results of this analysis, you can run the following code (after changing your current directory to wherever this README.md file is saved):

cd prompt_engineering/code
bash run_pipeline.sh

Questions

All questions should be directed to Matt DeVerna.