1. Which characterises inference rather than training?
2. Which four situations force a product team to reason below the model API layer?
3. Why are smaller models disproportionately cheaper to serve?
4. Roughly what round-trip latency does distance alone impose per 1,000 km in optical fibre?
5. A team moves inference nearer users but leaves the vector database on another continent. What happens?
6. What does streaming output change?
7. Why does sending less context improve two things at once?
8. How does AI accelerator capacity differ from ordinary cloud capacity?
9. Which terms are often more important to negotiate than unit price? Select all that apply.
10. Requests are growing faster than users. What is the most likely cause?