Language Models Can Order Political Positions Without Knowing Their Magnitude
November 18, 2026
Testing models on party positions in Japan and South Korea reveals that LLMs can often correctly rank entities by their political stances even if they cannot place those positions on an anchored scale of magnitude.
HOW THIS AFFECTS YOU
●
researcherWhen evaluating social knowledge in models, distinguish between their ability to rank items and their ability to quantify them.