1. Why can an AI feature not be specified the way conventional software can?
2. Which must come from a system of record rather than a generative model? Select all that apply.
3. What is the detectability test for an AI product surface?
4. Which responsibilities are new to an AI product manager? Select all that apply.
5. A medical triage classifier is worthless or harmful below a quality bar. What kind of product is it?
6. What does a manual prototype establish that building does not?
7. How should an interaction pattern be selected?
8. What is the highest-leverage design move in an AI product?
9. Why must refusal behaviour be in the evaluation set?
10. Which four sets make up an evaluation suite?
11. What characterises grading criteria that discriminate?
12. An offline score rises but nothing changes for users. Which explanations are plausible? Select all that apply.
13. What is the distinctive release risk in an AI product?
14. What is the central economic difference from conventional software?
15. Which cost lever is usually the largest single saving?
16. What are the five rungs of the capability ladder in order?
17. What test should precede moving up a rung?
18. Roughly what round-trip latency does distance impose per 1,000 km in fibre?
19. A team moves inference nearer users but leaves the vector index on another continent. What happens?
20. Why is sending a customer document to a model API a data transfer?
21. What is the architectural lesson of the EU–US transfer framework history?
22. Which component most commonly defeats a careful residency design?
23. What did analysis of GPT-3's training data find about language composition?
24. Which model-API failure has no error attached?
25. According to the IEA's Energy and AI analysis, what was global data centre electricity consumption in 2024 and what is projected for 2030?