Chapter 23 Quiz: Cloud AI Services and APIs

Chapter 23 Quiz: Cloud AI Services and APIs

Multiple Choice

Question 1. Which of the following best describes "data gravity" in the context of cloud AI strategy?

(a) The tendency for large datasets to require more compute resources for processing.
(b) The accumulation of data and dependencies in one cloud platform, making it progressively more costly to move workloads elsewhere.
(c) The gravitational pull of data centers located in urban areas rather than rural areas.
(d) The physical weight of storage hardware in on-premises data centers.

Question 2. A company's ML training workloads run in unpredictable bursts — heavy training for two weeks, then minimal usage for a month. Which pricing model would be most cost-effective for these training workloads?

(a) Reserved instances with a 3-year commitment
(b) On-demand instances
(c) Spot or preemptible instances with checkpointing
(d) Provisioned throughput

Question 3. An enterprise decides to use Azure OpenAI Service for LLM access, AWS SageMaker for custom ML models, and Google Vision AI for image analysis. Which cloud strategy does this represent?

(a) Single-cloud strategy
(b) Full multi-cloud strategy
(c) Hybrid cloud strategy
(d) Primary cloud with selective multi-cloud

Question 4. According to the chapter, which cost component of cloud AI total cost of ownership is most frequently underestimated?

(a) GPU compute costs for model training
(b) Cloud storage for training data
(c) Engineering time (salaries and benefits of ML engineers and data scientists)
(d) Cloud provider licensing fees

Question 5. What is model distillation in the context of LLM cost optimization?

(a) Removing unnecessary layers from a neural network to reduce its size
(b) Using a large, expensive model to generate training data for fine-tuning a smaller, cheaper model to approximate the larger model's performance
(c) Converting a cloud-hosted model to run on local hardware
(d) Compressing model weights using quantization to reduce memory requirements

Question 6. An organization's customer service chatbot processes 50,000 queries per day, with 70% being simple FAQ-type questions and 30% being complex inquiries requiring nuanced reasoning. Which cost optimization strategy would be most effective?

(a) Switching all queries to the cheapest available model
(b) Implementing tiered model routing — routing simple queries to a cheaper model and complex queries to a more capable model
(c) Caching all responses and serving cached answers for all repeat queries
(d) Batching all queries into hourly blocks for batch processing

Question 7. Which of the following is NOT one of the five strategic questions for cloud AI vendor selection described in the chapter?

(a) Where is our data already?
(b) What does our team know?
(c) Which provider has the most AI services?
(d) What does our security require?

Question 8. A healthcare company must ensure that patient data remains within the European Union. Which cloud capability addresses this requirement?

(a) Auto-scaling
(b) Data residency and sovereign cloud offerings
(c) Spot instance pricing
(d) Serverless inference

Question 9. What is the primary purpose of an API gateway in a multi-cloud AI architecture?

(a) To accelerate model training by distributing workloads across multiple clouds
(b) To provide a centralized control point for routing, security, monitoring, and cost tracking across AI services from multiple providers
(c) To store copies of all data in each cloud provider for redundancy
(d) To automatically select the cheapest cloud provider for each API call

Question 10. Athena Retail Group's anonymization pipeline for LLM queries performs the following steps in which order?

(a) Send query to LLM → detect PII → tokenize PII → return response
(b) Detect PII → tokenize PII → send anonymized query to LLM → de-tokenize PII in response
(c) Encrypt the entire query → send to LLM → decrypt response
(d) Remove all customer context → send generic query to LLM → add context to response

Question 11. Which cloud provider's AI strategy is most distinguished by its exclusive partnership with OpenAI?

(a) Amazon Web Services
(b) Microsoft Azure
(c) Google Cloud Platform
(d) IBM Cloud

Question 12. A company is evaluating cloud providers and discovers that its development team has four years of deep experience with AWS. According to the chapter, this "knowledge lock-in" represents which type of switching cost?

(a) A direct financial cost that can be calculated precisely
(b) An organizational cost measured in months of reduced productivity and retraining
(c) A contractual obligation that prevents switching
(d) A temporary inconvenience that resolves within days

Question 13. Which of the following represents a valid criticism of a full multi-cloud strategy (running all workloads on all cloud providers)?

(a) Multi-cloud eliminates all vendor lock-in.
(b) Multi-cloud reduces costs through competitive pricing across providers.
(c) Multi-cloud increases operational complexity and may result in teams knowing all platforms superficially rather than one deeply.
(d) Multi-cloud is only available to Fortune 500 companies.

Question 14. An organization currently spends $2 million annually on AWS cloud services. AWS offers a 3-year reserved capacity commitment with a 35% discount, requiring a minimum annual spend of $1.8 million. What is the total savings over three years if the organization's actual spend would have been $2 million per year at on-demand rates?

(a) $600,000
(b) $1,050,000
(c) $2,100,000
(d) Cannot be calculated without knowing the specific services

Question 15. According to the chapter, what is the relationship between compliance frameworks (SOC 2, HIPAA) and AI-specific security?

(a) Compliance frameworks fully address all AI security concerns.
(b) AI security is a subset of compliance — if you meet SOC 2, your AI is secure.
(c) Compliance frameworks provide a necessary baseline, but AI-specific risks (prompt injection, model inversion, training data leakage) require additional security practices.
(d) Compliance frameworks are irrelevant to AI workloads.

Short Answer

Question 16. Explain why data egress costs can become a dominant cost factor in multi-cloud AI architectures. Provide a specific example involving data transfer between two cloud providers.

Question 17. Ravi presents cloud AI costs to Athena's CFO as "cost per unit of business value" rather than raw cloud spend. Explain why this framing is more effective. Give two examples of cost-per-unit metrics that Ravi might use.

Question 18. A startup founder says: "We should build our AI on all three clouds from day one so we're never locked in." Drawing on the chapter, explain why this approach is likely to fail. What alternative would you recommend?

Question 19. Describe three cost optimization techniques specific to LLM API costs. For each technique, explain the mechanism by which it reduces cost and estimate the typical percentage reduction.

Question 20. The chapter states that "the best time to plan for exit is before you enter." What does this mean in the context of cloud vendor selection? What specific information should be documented at the time of vendor selection?

True or False (Explain Your Reasoning)

Question 21. True or False: "All three major cloud providers (AWS, Azure, GCP) offer equivalent AI and ML capabilities, so the choice between them is primarily a matter of pricing."

Question 22. True or False: "For most organizations, the cloud compute bill represents the majority of total cost of ownership for AI initiatives."

Question 23. True or False: "An organization that uses AWS SageMaker for model training and Azure OpenAI Service for LLM access is practicing a multi-cloud strategy."

Question 24. True or False: "Data anonymization before sending queries to an external LLM API eliminates all privacy risk associated with using that API."

Question 25. True or False: "Vendor lock-in is always a negative outcome that should be avoided in cloud AI strategy."

Answer key appears in Appendix B: Answers to Selected Exercises and Quizzes.