Prompting and Task Execution

• Topic 01

  Anatomy of an Effective Prompt

  45 questions
• Topic 02

  Providing Context and Source Material

  45 questions
• Topic 03

  Structuring Outputs

  48 questions
• Topic 04

  Multi-Step and Iterative Prompting

  45 questions
• Topic 05

  Few-Shot and Example-Driven Prompting

  45 questions
• Topic 06

  Common Prompting Pitfalls

  45 questions

A few questions from this unit, with the answer and a full explanation. The complete bank is available when you start practising.

1. A user notices that whichever request they put last in a long, multi-part prompt tends to be the one done least well. What does this best illustrate?

   • A
     The last request is always impossible to satisfy
   • B
     Adding more requests fixes the dilution
   • C
     The model can only ever do the first request
   • D
     Overloading dilutes attention across many asks, so some — often not all — are handled poorly
     Correct answer
   Explanation

   The pattern reflects overloading: with many competing asks, some get less attention and are handled poorly. The remedy is fewer asks per prompt, not reordering or padding. Key takeaway: patchy handling of a multi-part prompt is a symptom of overloading.

2. When a draft is mostly good but has one weak paragraph, which is generally the better choice?

   • A
     Accepting the weak paragraph unchanged
   • B
     A full rewrite of the entire draft
   • C
     A targeted follow-up that fixes the one paragraph
     Correct answer
   • D
     Discarding the draft and starting a new conversation
   Explanation

   A targeted follow-up addresses the single weak paragraph while keeping the good material, which is faster and lower-risk than a full rewrite. Rewriting everything risks losing the parts that already work. Key takeaway: fix the specific weak part with a follow-up rather than rewriting the whole draft.

3. A user must pick five examples to represent a classification task with three categories and some tricky mixed cases. Which selection is most likely to produce reliable, unbiased results?

   • A
     Examples that cover all three categories and include a tricky mixed case, balanced so no category is over-represented
     Correct answer
   • B
     Five examples chosen at random with no coverage check
   • C
     Five easy examples from just one category
   • D
     Five examples that all illustrate the same clear-cut case
   Explanation

   Reliable, unbiased few-shot results come from examples that cover every category, include the tricky cases, and are balanced so the set does not skew toward one class. A skewed set propagates bias into the classifications. Key takeaway: choose balanced, category-covering examples that include edge cases to avoid teaching a skewed pattern.