Aligning Large Language Models with Diverse Political Viewpoints

Dominik Stammbach, Philine Widmer, Eunjung Cho, Caglar Gulcehre, Elliott Ash
6/20/2024
cs.CL

No official code found. Best implementation: dominiksinsaarland/aligning-LLMs-with-political-views (1 star, last commit 2025-03-06).

Abstract

Large language models such as ChatGPT exhibit striking political biases. If users query them about political information, they often take a normative stance. To overcome this, we align LLMs with diverse political viewpoints from 100,000 comments written by candidates running for national parliament in Switzerland. Models aligned with this data can generate more accurate political viewpoints from Swiss parties, compared to commercial models such as ChatGPT. We also propose a procedure to generate balanced overviews summarizing multiple viewpoints using such models. The replication package contains all code and data.

View on arXivView PDF

Code Implementations(1)

dominiksinsaarland/aligning-LLMs-with-political-viewsCommunity100%
12PythonOct 2, 20241 years ago

Similar Papers

Multilingual Political Views of Large Language Models: Identification and Steering

87% match1 repo22025

Algorithmic Fidelity of Large Language Models in Generating Synthetic German Public Opinions: A Case Study

86% match1 repo22024

Large Language Models Reflect the Ideology of their Creators

85% match2 repos292024

GermanPartiesQA: Benchmarking Commercial Large Language Models and AI Companions for Political Alignment and Sycophancy

85% match2024

The Political Preferences of LLMs

85% match2024

Large Language Models' Detection of Political Orientation in Newspapers

84% match2024

Cite this paper

@article{stammbach2024aligning,
  title  = {Aligning Large Language Models with Diverse Political Viewpoints},
  author = {Dominik Stammbach and Philine Widmer and Eunjung Cho and Caglar Gulcehre and Elliott Ash},
  year   = {2024},
  eprint = {2406.14155},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2406.14155v2}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2406.14155.svg)](https://codeofpaper.com/paper/2406.14155)

Discussion