Privacy Notice: When you sign in with GitHub, your quiz attempts are stored in Google Firebase so you can track your progress. This data is only accessible to the course instructor, is not otherwise shared, and will be deleted at the end of the semester. For earlier deletion, contact the instructor. Prefer to stay anonymous? Use guest mode or a GitHub account without your real name.
Loading...
Sign in to track your progress and see your results across devices.
●Progress is being saved
02 Lecture 2 Quiz: From Copy-Paste to Crafted Prompts
Score:0 / 15
Question Overview
PendingCorrectIncorrect
Instructions
This is an interactive quiz. Sign in with GitHub to save your progress, or continue as a guest.
Select your answer for each question
Get immediate feedback on your choices
Retry questions as many times as you need
Track your progress across sessions
Question1
Three Approaches
In the lecture, three approaches to the Stack Overflow Challenge were compared. What is the main distinguishing factor between them?
✓Correct!
The three approaches (Copy-Paste, Structured Breakdown, Professional) differ primarily in how much engineering thinking and structure the developer applies before and during prompting. The same AI tool can produce vastly different results depending on the approach taken.
✗Not quite.
Hint
Think about what changes between the three approaches — the tools stay the same.
💡Solution
The three approaches (Copy-Paste, Structured Breakdown, Professional) differ primarily in how much engineering thinking and structure the developer applies before and during prompting. The same AI tool can produce vastly different results depending on the approach taken.
Question2
Prompting Pitfalls
What does the “80-20 problem” refer to in the context of AI-assisted development?
✓Correct!
The 80-20 problem describes how a single broad prompt can produce code that appears to work within minutes, but fixing the remaining issues — debugging, data access problems, subtle bugs — takes hours. The initial speed is seductive but misleading. While distractor A sounds plausible, the lecture specifically defines the 80-20 problem as the gap between “looks done” and “actually works.”
✗Not quite.
Hint
Consider the relationship between the speed of the first result and the time spent debugging.
💡Solution
The 80-20 problem describes how a single broad prompt can produce code that appears to work within minutes, but fixing the remaining issues — debugging, data access problems, subtle bugs — takes hours. The initial speed is seductive but misleading. While distractor A sounds plausible, the lecture specifically defines the 80-20 problem as the gap between “looks done” and “actually works.”
Question3
Hallucinations
According to the lecture, what increases the “hallucination surface area” when prompting an AI?
✓Correct!
A single, broad prompt gives the AI maximum freedom. With no constraints or structure, the model fills gaps by generating plausible-sounding content. The broader the prompt, the more room for hallucination. Focused, specific prompts with clear requirements reduce this risk. Specifying a library (B) actually constrains the AI, which reduces hallucination surface area.
✗Not quite.
Hint
Think about what gives the AI more freedom to invent things.
💡Solution
A single, broad prompt gives the AI maximum freedom. With no constraints or structure, the model fills gaps by generating plausible-sounding content. The broader the prompt, the more room for hallucination. Focused, specific prompts with clear requirements reduce this risk. Specifying a library (B) actually constrains the AI, which reduces hallucination surface area.
Question4
Prompting Best Practices
Why does the lecture recommend writing prompts in English, even for German-speaking students?
✓Correct!
Research shows that LLMs are trained predominantly on English text (GPT-3: ~93%, LLaMA: ~90%). Studies by Shi et al. (2022), Etxaniz et al. (2023), and Huang et al. (2023) demonstrate that prompting in English — even when the question is in another language — consistently improves reasoning performance. The model can still output in any language you request. While English is common in code (A), the recommendation is about reasoning quality, not language matching.
✗Not quite.
Hint
Consider the composition of the training data.
💡Solution
Research shows that LLMs are trained predominantly on English text (GPT-3: ~93%, LLaMA: ~90%). Studies by Shi et al. (2022), Etxaniz et al. (2023), and Huang et al. (2023) demonstrate that prompting in English — even when the question is in another language — consistently improves reasoning performance. The model can still output in any language you request. While English is common in code (A), the recommendation is about reasoning quality, not language matching.
Question5
Structured Breakdown
In Approach 2 (Structured Breakdown), what is the purpose of the first prompt?
✓Correct!
Approach 2 starts with a research prompt that asks the AI to analyze and compare data source options — their pros, cons, rate limits, and access requirements. No code is requested yet. This strategy of using the AI as an analyst before using it as a coder consistently produces better results.
✗Not quite.
Hint
What do you do before you start building anything?
💡Solution
Approach 2 starts with a research prompt that asks the AI to analyze and compare data source options — their pros, cons, rate limits, and access requirements. No code is requested yet. This strategy of using the AI as an analyst before using it as a coder consistently produces better results.
Question6
Prompt Design
Why does breaking a task into multiple focused prompts (instead of one big prompt) improve AI output quality?
✓Correct!
When each prompt targets a single, well-defined task, the AI has a clear scope with less room to hallucinate. Additionally, if something goes wrong, the failure is isolated to one module rather than tangled across the entire codebase. This mirrors the software engineering principle of separation of concerns.
✗Not quite.
Hint
Think about what happens when a bug appears in a monolithic script vs. separate modules.
💡Solution
When each prompt targets a single, well-defined task, the AI has a clear scope with less room to hallucinate. Additionally, if something goes wrong, the failure is isolated to one module rather than tangled across the entire codebase. This mirrors the software engineering principle of separation of concerns.
Question7
Professional Approach
In Approach 3 (Professional), why are tests written before the implementation code?
✓Correct!
Writing tests first (TDD-style) creates a specification that the implementation must satisfy. If the AI hallucinates an API endpoint or generates buggy code, the tests catch it immediately — there are no silent bugs. Tests serve as automatic hallucination detectors.
✗Not quite.
Hint
What happens when the AI generates incorrect code but tests already define the expected behavior?
💡Solution
Writing tests first (TDD-style) creates a specification that the implementation must satisfy. If the AI hallucinates an API endpoint or generates buggy code, the tests catch it immediately — there are no silent bugs. Tests serve as automatic hallucination detectors.
Question8
Engineering Judgment
The lecture gives an example of typing 6 milestone dates manually instead of building an AI-powered scraper. What principle does this illustrate?
✓Correct!
When you already have a small, known dataset (6 milestones), typing it manually takes 2 minutes versus 20+ minutes to build an automated solution that might hallucinate dates. Good engineering judgment means choosing the right tool — including no tool at all. The point is not that automation is bad (D), but that the cost-benefit analysis matters.
✗Not quite.
Hint
Consider the time trade-off between automation and manual work for small tasks.
💡Solution
When you already have a small, known dataset (6 milestones), typing it manually takes 2 minutes versus 20+ minutes to build an automated solution that might hallucinate dates. Good engineering judgment means choosing the right tool — including no tool at all. The point is not that automation is bad (D), but that the cost-benefit analysis matters.
Question9
AI Diary
Which of the following is NOT a stated purpose of the AI Diary in this course?
✓Correct!
The AI Diary serves three purposes: self-reflection on AI-assisted development skills, providing data for later coursework analysis, and creating professional evidence of your workflow. It does not replace git commit messages — in fact, Approach 3 uses both diary entries AND meaningful commit messages together.
✗Not quite.
Hint
Think about what the diary tracks that git commits do not — and vice versa.
💡Solution
The AI Diary serves three purposes: self-reflection on AI-assisted development skills, providing data for later coursework analysis, and creating professional evidence of your workflow. It does not replace git commit messages — in fact, Approach 3 uses both diary entries AND meaningful commit messages together.
Question10
AI Diary
Why is the built-in chat history of tools like Copilot or ChatGPT insufficient for tracking AI interactions?
✓Correct!
Built-in chat history has significant gaps: no correspondence between prompts and code changes, no tracking of how many iterations a task required, no record of intent versus actual request, no way to compare across different tools, and fragile storage that can be lost when the IDE restarts.
✗Not quite.
Hint
Think about what information you lose when you only have a chat transcript.
💡Solution
Built-in chat history has significant gaps: no correspondence between prompts and code changes, no tracking of how many iterations a task required, no record of intent versus actual request, no way to compare across different tools, and fragile storage that can be lost when the IDE restarts.
Question11
Tool Configuration
What problem do project instruction files like CLAUDE.md and .github/copilot-instructions.md solve?
✓Correct!
The AI does not remember project conventions across sessions. Without instruction files, you would need to re-specify conventions (diary format, commit message style, testing requirements, coding standards) at the start of every new session. Instruction files encode these conventions once and are read automatically.
✗Not quite.
Hint
What happens when you start a new chat session — what does the AI forget?
💡Solution
The AI does not remember project conventions across sessions. Without instruction files, you would need to re-specify conventions (diary format, commit message style, testing requirements, coding standards) at the start of every new session. Instruction files encode these conventions once and are read automatically.
Question12
Tool Comparison
Which of the following is a key difference between GitHub Copilot and Claude Code in VS Code?
✓Correct!
Both tools can read/write files and run terminal commands in agent mode. The key differences are: Copilot offers model choice (GPT-4o, Claude Sonnet, Gemini), while Claude Code uses Anthropic models with up to 1M token context windows. Both have free tiers available for students.
✗Not quite.
Hint
Both tools can do similar things — what differs is the models and context size.
💡Solution
Both tools can read/write files and run terminal commands in agent mode. The key differences are: Copilot offers model choice (GPT-4o, Claude Sonnet, Gemini), while Claude Code uses Anthropic models with up to 1M token context windows. Both have free tiers available for students.
Question13
Tool Usage
What is the difference between “Ask” mode and “Agent/Edit” mode in AI coding tools?
✓Correct!
In Ask mode, the AI acts as a consultant — it answers questions, explains code, and suggests changes, but does not edit files directly. In Agent/Edit mode, the AI acts as a collaborator — it can read and write files, run commands, and make multi-file changes while you review the results.
✗Not quite.
Hint
Think about the analogy: consultant vs. collaborator.
💡Solution
In Ask mode, the AI acts as a consultant — it answers questions, explains code, and suggests changes, but does not edit files directly. In Agent/Edit mode, the AI acts as a collaborator — it can read and write files, run commands, and make multi-file changes while you review the results.
Question14
Advanced Prompting
The lecture discusses encoding a multi-step process into a single prompt (a “meta-prompt”). What is the main trade-off?
✓Correct!
A single meta-prompt that encodes the decomposition process is more convenient (one prompt instead of five), but the AI decides the module boundaries, data source, and library choices. When you decompose manually, you build the mental model of the solution — essential for debugging when something breaks.
✗Not quite.
Hint
Who has the mental model of the solution when the AI does the decomposition?
💡Solution
A single meta-prompt that encodes the decomposition process is more convenient (one prompt instead of five), but the AI decides the module boundaries, data source, and library choices. When you decompose manually, you build the mental model of the solution — essential for debugging when something breaks.
Question15
Prompt Design
In Approach 3, Prompt 0 includes the instruction “Do NOT implement any logic yet.” Why is this important?
✓Correct!
Controlling what the AI does NOT do is just as important as telling it what to do. Without the explicit constraint, the AI will generate implementation code in every file, bypassing the planned workflow of writing tests first, then implementing against them.
✗Not quite.
Hint
What would happen if you did not include this constraint?
💡Solution
Controlling what the AI does NOT do is just as important as telling it what to do. Without the explicit constraint, the AI will generate implementation code in every file, bypassing the planned workflow of writing tests first, then implementing against them.