LLM Bias Grading Turns Social Science Into The Test Case

What this establishes institutionally is a new burden of proof for AI-assisted political-text analysis. Work summarized by the American Journal of Political Science argues that LLMs can locate policy positions from political texts better than experts, while CEA-List's work treats political bias in LLMs as something that varies by language and model size through a . The Manzi paper sits between those claims: it relies on model judgment while inviting scrutiny of model politics. The consequence is immediate for computational social science: ideological measurement now requires a…