Have we been measuring AI political bias wrong? A better approach is possible (opens in new tab)

Why ideological preferences and epistemic failure in LLMs are not the same thing — and why the difference matters