personallm's Introduction

PersonaLLM: Investigating the Ability of Large Language Models to Express Personality Traits

This repo contains the official code for running LLM persona experiments and subsequent analyses in the PersonaLLM paper.

Simulate `LLM personas`

We first create 10 personas for each of 32 personality types.

conda activate audiencenlp
python3.9 run_bfi.py --model "GPT-3.5-turbo-0613"
python3.9 run_bfi.py --model "GPT-4-0613"
python3.9 run_bfi.py --model "llama-2"

Generate stories with `LLM personas`

python3.9 run_creative_writing.py --model "GPT-3.5-turbo-0613"
python3.9 run_creative_writing.py --model "GPT-4-0613"
python3.9 run_creative_writing.py --model "llama-2"

References

If you use this repository in your research, please kindly cite our paper:

@inproceedings{jiang-etal-2024-personallm,
    title = "{P}ersona{LLM}: Investigating the Ability of Large Language Models to Express Personality Traits",
    author = "Jiang, Hang  and
      Zhang, Xiajie  and
      Cao, Xubo  and
      Breazeal, Cynthia  and
      Roy, Deb  and
      Kabbara, Jad",
    editor = "Duh, Kevin  and
      Gomez, Helena  and
      Bethard, Steven",
    booktitle = "Findings of the Association for Computational Linguistics: NAACL 2024",
    month = jun,
    year = "2024",
    address = "Mexico City, Mexico",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.findings-naacl.229",
    pages = "3605--3627",
}

Acknowledgement

PersonaLLM is a research program from MIT Center for Constructive Communication (@mit-ccc), MIT Media Lab, and Stanford University. We are interested in drawing from social and cognitive sciences to understand the behaviors of foundation models.

Recommend Projects