AIMentor — AI Expertise Learning Platform

Question 1easy

What is the core principle of effective prompting?

Make prompts as short as possibleBe specific — pin down task, constraints, format, and audienceAlways use the largest modelRepeat the instruction many times

Question 2easy

What does 'few-shot' prompting mean?

Making only a few API callsIncluding a few input→output examples in the promptUsing a small modelLimiting the output to a few tokens

Question 3medium

When is chain-of-thought ('think step by step') most useful?

For simple lookupsFor multi-step reasoning tasks (math, logic, complex analysis)To reduce token usageTo make output deterministic

Question 4medium

You need output that code will parse. Prompt-specified format vs. schema-enforced structured output — which is stronger?

Prompt-specified format is a guaranteeSchema-enforced structured output is a guarantee; a prompt-specified format is only a requestThey are identicalNeither can produce JSON

Question 5hard

What is prompt injection?

Injecting extra tokens to speed up generationUntrusted input containing instructions that hijack the model's behaviorA way to compress the promptAdding few-shot examples

Question 6medium

What's a good first-line defense when a prompt must include untrusted input?

Trust the model to figure it outWrap the input in clear delimiters and instruct the model to treat it as data, not instructionsDelete all punctuation from the inputUse a higher temperature

Question 7medium

What's the right way to iterate on a prompt?

Change many things at once and hope it's betterChange one element at a time and re-test on a small example setOnly change the model, never the promptTrust your intuition without testing

Question 8hard

A prompt heavily tuned for an older model behaves oddly (over-triggers a tool) on a newer model. Why?

Newer models ignore promptsNewer models follow instructions more literally, so aggressive 'CRITICAL: YOU MUST' language over-triggersThe prompt is too shortTemperature changed automatically

Question 9medium

For teaching a model a specific output style, which tends to work best?

A long list of prohibitions ('never do X, never do Y')Positive examples showing the desired outputRaising max_tokensLowering the temperature to 0

Question 10medium

Which elements make a prompt more specific and reliable? (Select all that apply)

Explicit output formatTarget audienceLength/constraint limitsA higher temperature

Select all that apply.

Question 11easy

What is the role/persona (system prompt) mainly for?

Storing the API keySetting tone, vocabulary, and standards for the whole conversationCounting tokensEnabling streaming

Question 12hard

Why keep a small set of input→expected examples while developing a prompt?

To increase the token countTo measure whether a prompt change actually improved results (a mini eval set)Because the API requires examplesTo slow down iteration deliberately

Question 13medium

Term (hyphenated or not) for including example input→output pairs in a prompt to teach format/behavior.

Question 14hard

Two words: the attack where untrusted input carries instructions that hijack the model (e.g. 'ignore previous instructions').

Prompt Engineering — Practice quiz