AIF-C01 Study Platform

Lesson 8 Quiz — Prompt Engineering Techniques & Risks (10 easy + 10 hard)

20 questions.

Q1easymcq

What is a 'negative prompt'?

Q2easymcq

A prompt with zero examples before the real question is called:

Q3easymcq

A prompt with several worked examples before the real question is called:

Q4easymcq

Instructing a model to 'think step by step' before answering is an example of:

Q5easymcq

A reusable prompt structure with placeholders swapped in per request is called a:

Q6easymcq

An attacker crafting input specifically to bypass a model's built-in safety restrictions is called:

Q7easymcq

An attacker's input redirecting the model away from its intended task toward the attacker's goal is called:

Q8easymcq

Which AWS feature lets you create, version, and test prompt templates in a managed way?

Q9easymcq

'Poisoning' in prompt engineering risk refers to:

Q10easymcq

Which best practice pairs prompting with actual safety controls rather than relying on wording alone?

Q11hardmcq

A prompt provides exactly one worked example before the real task. What technique is this?

Q12hardmulti

Which of the following are prompt-related risks covered in this lesson?

Select all that apply.

Q13hardmcq

A malicious actor gets a coding assistant to ignore its task and instead reveal its hidden system prompt. This is best classified as:

Q14hardmcq

A user crafts input specifically designed to make a model generate content it's built to refuse (e.g., harmful instructions). This is best classified as:

Q15hardordering

Order these prompting approaches from least to most guidance provided to the model.

1.Chain-of-thought with few-shot examples
2.Few-shot
3.Zero-shot
4.Single-shot
Q16hardmatching

Match each prompt risk to its description.

Exposure
Poisoning
Hijacking
Jailbreaking
Q17hardmcq

A RAG pipeline retrieves a document that was deliberately edited by an attacker to contain hidden instructions manipulating the model's answer. This is an example of:

Q18hardmcq

A team keeps rewriting the same prompt inconsistently across different parts of their application, causing inconsistent output. Which AWS feature directly addresses this?

Q19hardmulti

Which of these are prompt engineering best practices from this lesson?

Select all that apply.

Q20hardmcq

A math word-problem prompt gets noticeably more accurate answers after adding 'explain your reasoning step by step' to the instruction. This demonstrates the value of: