Ideological Alignment in LLMs

Definition

The emergent political, philosophical, and normative biases embedded within Large Language Models through pretraining corpora, RLHF preference datasets, and system prompt constraints.

Key Characteristics

• Multidimensional positioning across economic and social governance axes.
• Sensitivity to prompt persona and framing permutations.
• Differential alignment between open-weight and commercial API models.

Applications

• AI governance auditing
• Model alignment research
• Curricular bias detection

Mentions in Source

• “RLHF and system prompting induce measurable ideological shifts across models.” — ai-political-compass