Skip to content

Chapter 12 Quiz

← Back to chapter

1. According to the chapter, how do large language models generate their responses?

  1. By retrieving verified facts from a curated database
  2. By predicting the most likely sequence of words based on patterns learned from training data
  3. By reasoning through problems using logical deduction
  4. By searching the internet in real time for current information
Answer

B. LLMs work by learning statistical patterns in language during training on billions of sentences. When given a prompt, the model generates a response by predicting the most likely sequence of words that should follow, given all the patterns it has learned. It does not "understand" the question the way a human does -- it produces text that is statistically consistent with its training data. This is why LLMs can produce confident but factually incorrect text.

Concept tested: Large Language Models

Bloom's level: Understand

2. What does the CRAFT framework for prompt engineering stand for?

  1. Create, Review, Analyze, Finalize, Test
  2. Context, Role, Audience, Format, Tone
  3. Clarity, Relevance, Accuracy, Flow, Transparency
  4. Command, Refine, Assess, Filter, Transform
Answer

B. The CRAFT framework provides a repeatable structure for building effective prompts: Context (background information), Role (who the AI should write as), Audience (who will read the output), Format (structure of the output), and Tone (style of writing). This framework applies communication principles -- audience awareness, purpose clarity, structure -- to the act of instructing an AI.

Concept tested: Prompt Engineering

Bloom's level: Remember

3. In the human-AI collaboration workflow described in the chapter, what is the final step?

  1. AI generates the final polished version
  2. The human sends the AI output directly to the audience
  3. The human approves the final version and takes responsibility
  4. AI fact-checks its own output for accuracy
Answer

C. The human-AI collaboration cycle ends with the human approving the final version. The key principle is that the human is always in the loop and always accountable for the final output. An AI-generated report with your name on it is your report -- if it contains errors, you are responsible, not the AI. The five steps are: human defines the goal, AI generates a draft, human reviews and refines, AI assists with polish, and human approves the final version.

Concept tested: Human-AI Collaboration

Bloom's level: Understand

4. Which of the following is an example of the AI limitation called "hallucination"?

  1. An LLM refuses to answer a question about a topic released after its training cutoff
  2. An LLM generates a citation to a research paper that does not actually exist
  3. An LLM produces a response that is too long for the requested format
  4. An LLM gives different answers to the same prompt on different occasions
Answer

B. Hallucination occurs when LLMs generate text that is factually incorrect but sounds authoritative. The chapter specifically mentions that LLMs may cite papers that do not exist, invent statistics, or describe processes that do not work. This happens because the model is generating plausible-sounding text, not retrieving verified facts. Option A describes the knowledge cutoff limitation, not hallucination.

Concept tested: AI Limitations

Bloom's level: Apply

5. What are the two types of AI summarization described in the chapter?

  1. Manual summarization and automated summarization
  2. Brief summarization and detailed summarization
  3. Extractive summarization and abstractive summarization
  4. Technical summarization and executive summarization
Answer

C. The chapter describes two types: extractive summarization, where the AI selects and combines the most important sentences from the original text using the author's own words, and abstractive summarization, where the AI generates new sentences that capture the meaning of the original text as a paraphrase. Both types carry the risk of omitting important nuances or introducing errors.

Concept tested: AI Summarization

Bloom's level: Remember

6. According to the chapter, what guiding principle should you use when navigating AI ethics in communication?

  1. Always disclose AI use in every communication without exception
  2. AI-generated content should never be used in professional settings
  3. Would you be comfortable if your audience knew exactly how this content was produced?
  4. As long as the output is accurate, the process of creating it does not matter
Answer

C. The chapter states that a reliable guiding principle for AI ethics is: "Would you be comfortable if your audience knew exactly how this content was produced?" If the answer is yes, you are probably on solid ethical ground. If the answer is no, reconsider your approach. This principle is more nuanced than a blanket rule because norms around AI disclosure are still evolving across different contexts.

Concept tested: AI Ethics Communication

Bloom's level: Understand

7. An engineer uses an AI tool to draft a blog post and publishes it without reviewing the content. The post contains a fabricated statistic. According to the chapter, who is responsible for this error?

  1. The AI tool provider for generating incorrect information
  2. The engineer who published the content without fact-checking
  3. The audience for not verifying the information themselves
  4. No one, because AI-generated errors are unavoidable
Answer

B. The chapter is explicit that the human is always accountable for the final output. An AI has no professional reputation, no career at stake, and no ethical obligations -- it cannot take responsibility for errors. Fact-checking AI output is described as a professional obligation, not an optional step. The engineer who publishes AI-generated content without review is responsible for any errors it contains.

Concept tested: Fact-Checking AI Output

Bloom's level: Evaluate

8. Which phase of AI workflow integration involves using AI to flag sections of documentation that may be outdated?

  1. Planning phase
  2. Drafting phase
  3. Review phase
  4. Maintenance phase
Answer

D. The chapter describes five phases of AI workflow integration. The maintenance phase involves using AI to flag sections of documentation that may be outdated based on new product releases or API changes. The planning phase uses AI for brainstorming and research; the drafting phase for generating first drafts; the review phase for grammar and readability checking; and the publishing phase for generating summaries and abstracts.

Concept tested: AI Workflow Integration

Bloom's level: Remember

9. You are tasked with using an AI writing assistant to draft a technical design document for your team. Using the CRAFT framework, write a complete prompt that would produce a useful first draft. Explain why each element of your prompt matters.

Answer

A strong answer should demonstrate all five CRAFT elements with clear reasoning. For example:

Prompt: "You are a senior software engineer at a cloud infrastructure company (Role). Our team needs to migrate our user authentication service from a monolithic architecture to a microservices-based design to support our growing user base of 500K daily active users (Context). Write a 1,500-word design document for a technical audience of backend engineers and the engineering manager who will approve the project (Audience). Structure the document with sections for Problem Statement, Proposed Architecture, Migration Plan, Risk Assessment, and Timeline (Format). Use a professional, direct tone with specific technical details and avoid marketing language (Tone)."

Why each element matters:

  • Context gives the AI the background needed to generate relevant content rather than generic boilerplate
  • Role shapes the expertise level and perspective of the output
  • Audience determines vocabulary depth, technical detail level, and what to emphasize
  • Format ensures the output matches the structure the team expects
  • Tone prevents the AI from defaulting to an inappropriate style

Without these elements, the AI would produce a vague, generic draft that requires extensive revision. The CRAFT framework applies the same communication principles (audience awareness, purpose clarity, structure) to instructing the AI.

Concept tested: Prompt Engineering

Bloom's level: Create

10. The chapter describes several AI limitations including hallucination, knowledge cutoff, lack of reasoning, context limitations, and bias. Choose two of these limitations and explain how each one could specifically harm an engineer's credibility if the engineer published AI-generated content without proper review.

Answer

A strong answer should select two limitations and provide concrete, specific scenarios.

Hallucination: An engineer publishes a technical blog post where the AI cites a non-existent research paper to support a performance claim (e.g., "According to Smith et al. (2024) in IEEE Transactions..."). When readers try to find the cited paper, they discover it does not exist. This destroys the engineer's credibility because it suggests either dishonesty or a lack of due diligence. Readers will question every other claim in the post and in future publications.

Knowledge cutoff: An engineer uses AI to write documentation recommending a specific library version, but the AI's training data predates a critical security vulnerability disclosed in that version. Publishing this recommendation without checking current advisories could lead other developers to adopt vulnerable software, damaging the engineer's reputation and potentially causing security incidents.

Bias: An engineer uses AI to draft job descriptions or team communication, and the output contains subtly exclusionary language that reflects biases in the training data. Publishing this without review could harm hiring diversity and damage the engineer's standing as an inclusive team leader.

Lack of reasoning: An engineer publishes AI-generated performance benchmarks that contain mathematical errors in the calculations. The AI produced numbers that looked plausible but were arithmetically incorrect, and the engineer did not verify them.

In all cases, the core lesson is the same: fact-checking AI output is a professional obligation because the engineer's name -- not the AI's -- is attached to the published work.

Concept tested: AI Limitations, Fact-Checking AI Output

Bloom's level: Evaluate