1. What is the single most useful question to ask when someone proposes an AI project?
2. Why is labelling often the most expensive part of a machine learning project?
3. Which statement about a model's output is accurate?
4. What does it mean that "the answer must be in the input"?
5. Google Flu Trends over-predicted flu in 100 of 108 weeks. What went wrong?
6. How do AI, machine learning, neural networks and deep learning relate?
7. What distinguishes data science from machine learning as a project type?
8. Why do large language models produce confident falsehoods?
9. Applying the one-second rule, which is most likely automatable today?
10. A team delivers an accurate churn model but churn does not fall. Why?
11. Which are structural traits of a company genuinely built around AI? Select all that apply.
12. Why does a machine learning project loop rather than proceed linearly?
13. What begins, rather than ends, when a model is deployed? Select all that apply.
14. Netflix never deployed the algorithm that won its $1 million prize. What is the lesson?
15. Why should stakeholders specify outcomes rather than methods?
16. For a fraud model where one transaction in a thousand is fraudulent, why is overall accuracy a poor criterion?
17. What is shadow mode?
18. Three chained machine learning steps are each 95% accurate. What should you expect end to end?
19. What is the long tail, and why does it matter commercially?
20. Sampling 100 real cases yields 40 straightforward, 40 awkward, 20 genuinely hard. What does this indicate?
21. Why put a working pilot before writing an AI strategy?
22. Zillow Offers closed in November 2021 with over $500 million in losses after doubling home purchases twice that year. Which pitfall is this?
23. Why does removing gender and ethnicity from a model's inputs fail to prevent discrimination?
24. Researchers made a stop sign read as "Speed Limit 45" using black and white stickers, succeeding in 100% of lab images and 84.8% of drive-by frames. Why is this significant?
25. What is the most defensible position on AI and employment?