引導式測驗 · 中等 等級

西方中心偏見在大型語言模型測驗

測驗如何衡量文化一致性、基於調查的研究發現了什麼,以及為什麼全國平均和英語樣本限制泛化。

問題 1 的 8

What does “Western-centric bias” mean in this guide?

預覽每個問題8 問題

當您參加測驗時,就會出現答案和解釋。

  1. What does “Western-centric bias” mean in this guide?
    • A claim that all Western users share one opinion
    • A measured mismatch between model outputs and some non-Western or local reference values
    • A model’s refusal to answer every English prompt
    • A legal category in international law
  2. What source did the 2024 PNAS Nexus study use to compare cultural values?
    • A single social-media hashtag
    • A legal code from one country
    • The model’s own self-description
    • World Values Survey items and responses
  3. How many models did the 2024 study examine?
    • Two chatbots only
    • Twenty-one speech systems
    • Five GPT models
    • Every model available worldwide
  4. Which broad tendency did the 2024 study report in its World Values Survey comparison?
    • The study found no variation across questions
    • Responses leaned toward self-expression values associated with some Western settings in its framework
    • Every model matched every country’s survey results exactly
    • Models always preferred traditional authority values
  5. What did the 2026 moral-values study compare?
    • Model estimates compared with U.S. census responses only
    • A comparison of model responses across language-count benchmarks
    • Survey data from 48 countries without model-generated estimates
    • Model-generated moral-value estimates with survey data from 48 countries
  6. Why is an English-language survey benchmark a limitation for cultural analysis?
    • English respondents may be more educated or globally connected than the broader population
    • An English-language sample is representative of every country by default
    • Country coverage means all language groups were sampled equally
    • English proficiency has no relationship to education or global connectivity
  7. Does a country’s average survey response describe every person in that country?
    • Yes, if a model repeats it
    • Yes, every resident shares the mean
    • No; a national aggregate is not an individual-level preference
    • No, because surveys cannot be used
  8. Which evaluation practice improves a culturally specific product review?
    • Test relevant languages and contexts with local speakers and qualified sources
    • Assume translation guarantees cultural fit
    • Rely on one U.S. English benchmark
    • Prompt the model to be “neutral” once