On Generalization across Measurement Systems: LLMs Entail More Test-Time Compute for Underrepresented Cultures

Open in new window