LLMs show bias favoring left-leaning users in online political debates

A new study published on arXiv reveals that Large Language Models (LLMs) exhibit a bias towards left-leaning users in online political debates. Researchers conducted nearly 150,000 controlled multi-turn conversations, varying user attributes like political orientation, expertise, and rhetorical strategy. The findings indicate that LLMs are systematically more receptive to left-leaning participants, an asymmetry that persists even when users with different political leanings defend the same position or engage in the same conversation. While repeated interaction and expertise influence receptiveness, fine-tuning LLMs on different political corpora can alter or even reverse this observed bias. AI

IMPACT Highlights potential biases in LLM interactions that could affect online discourse and public opinion.

RANK_REASON The cluster contains a research paper published on arXiv detailing findings about LLM behavior. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Gotit.pub
• Hugging Face
• Large Language Models
• left-leaning users
• right-leaning users
• ScienceCast

• paper
• safety

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