LLMs show centrist, status-quo bias on concrete policy votes

A new study published on arXiv challenges the notion that large language models (LLMs) exhibit a consistent left-leaning political bias. Researchers found that while LLMs align with established findings when answering abstract political questionnaires, their voting patterns on concrete policy decisions in Switzerland shift towards the center. The study also revealed significant cross-linguistic inconsistencies in LLM responses and identified two models that consistently vote against change, regardless of the policy's direction. AI

IMPACT Challenges assumptions about LLM political alignment, suggesting a more nuanced, context-dependent behavior on policy issues.

RANK_REASON The cluster contains an academic paper detailing novel research findings on LLM behavior. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• arXiv
• GPT-5.4
• Gruene
• LLMs

• paper
• opinion

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →