Political Bias in Large Language Models within the context of the 2024 European Parliament elections
As a result of a research to Assess Political Bias in Large Language Models using Wahl-O-Mat questionnaire, to determine which political party is the most aligned for the respective LLM, seems the larger models, such as Llama3–70B, tend to align more closely with left-leaning political parties like GRUNE and Volt, while smaller models, often remain neutral, particularly in English language.
if you are not a Medium premium member, you can read it for free here.
Wahl-o-Mat
A digital tool designed to help voters in Germany decide which political party aligns most closely with their views. It’s typically released ahead of major elections. The user is presented with a series of political statements, ranging from social issues to economic policies, and cover a wide range of topics relevant to the political landscape at the time. For each statement, the user can decide whether they agree, are neutral, or disagree. In the end, the Wahl-o-mat compares the user’s responses with the official positions of the political parties participating in the election.
LLM Models
This research used the most popular text generation models of the Hugging Face platform, which you can see…
