LLMs Show Persistent Bias Toward Japanese Culture Across Languages, Raising Training Data Questions

A viral analysis shows that major LLMs default to Japanese cultural references when answering questions in non-native contexts — a pattern that persists across models and languages, suggesting systematic training data skew.