1. What have published multilingual benchmarks documented about model performance across languages?
2. What did analysis of GPT-3's training data find about its language composition?
3. What three things does the tokenisation disparity affect simultaneously?
4. Why is a translated evaluation set weak evidence of readiness in a new language?
5. A product answers in French from a knowledge base that exists only in English. What breaks?
6. Which costs of multi-region operation are most often missing from engineering plans?
7. Why should model availability be checked in every region before committing to a feature?
8. A product serves 80% of traffic from one region and 20% from another. Why can a total failure in the smaller region go unnoticed?
9. Why can an AI product not price regionally as freely as conventional software?
10. Which responses to price sensitivity in a lower-income market are viable? Select all that apply.