Task 4.1: Craft effective prompts

• AZero-shot prompting

  Incorrect

  Examples were supplied here, so the prompt isn't zero-shot.

• BRole prompting

  Incorrect

  No expert role was assigned in the prompt.

• CFew-shot prompting

  Correct

  Presenting several examples is few-shot prompting, which sits between zero-shot unpredictability and the precision of fine-tuning.

• DChain prompting

  Incorrect

  Everything was given in a single prompt, not built up over several turns.

Few-shot prompting gives several examples so that Copilot understands the patterns expected, which helps with sophisticated implementations and edge cases. GitHub's prompt engineering guide also recommends giving examples of input data and output.

• AKeep the order but repeat each rule twice so that Copilot cannot miss any of them

  Incorrect

  Repetition is not the guidance; ordering and clarity are.

• BMove the goal to the end, because Copilot reads the last sentence as the task

  Incorrect

  The guidance is to start with the broad goal, not end with it.

• CSplit it into one message per rule and send them all before stating the goal

  Incorrect

  Breaking down a task helps, but stating the goal after the rules keeps the ambiguity.

• DOpen with the goal, then list requirements with the memory rule first

  Correct

  Start general, then get specific, and put the most important requirements first; Copilot gives extra weight to the beginning and end of a prompt.

GitHub's prompt engineering guidance says to first give a broad description of the goal or scenario, then list specific requirements with the most important ones first. Copilot gives extra weight to instructions at the beginning and end of a prompt, so a critical rule buried in the middle is easier to lose.

• AOpen the files relevant to the new class and close the unrelated ones

  Correct

  For code completions, GitHub advises opening relevant files and closing irrelevant ones, because Copilot uses open files as context.

• BPin the prototype files so that Copilot treats them as lower priority

  Incorrect

  Pinning is not a documented way to weight context.

• CRestart the IDE so that Copilot forgets the project's existing code

  Incorrect

  A restart does not change how Copilot uses open files.

• DTurn on next edit suggestions so that only the current file is used

  Incorrect

  Next edit suggestions predict edits; they don't narrow context to one file.

Copilot uses the context available to it. For code completions in your IDE, open any relevant files and close irrelevant ones, because Copilot uses your open files to understand your request. In chat, highlight code and attach context with @ or variables.

• AZero-shot prompting

  Correct

  Zero-shot means Copilot generates code without specific examples, relying on its foundational training.

• BOne-shot prompting

  Incorrect

  No example was given in the comment.

• CFew-shot prompting

  Incorrect

  No examples were given in the comment.

• DRole prompting

  Incorrect

  No expert role was assigned.

In zero-shot prompting, Copilot generates code without any specific examples, relying solely on its foundational training, which suits common patterns such as a temperature conversion.

• AAsk the same question again until the right block is explained

  Incorrect

  Repeating the ambiguous question doesn't remove the ambiguity.

• BAdd "please" and "thank you" so the request is clearer

  Incorrect

  Politeness doesn't identify which code is meant.

• CName it: "What does the createUser function do?"

  Correct

  GitHub's guidance is to avoid ambiguous terms like "this" and name the function or response you mean.

• DAsk Copilot to explain every code block in the conversation

  Incorrect

  That avoids the question instead of making it specific.

Avoid ambiguous terms. "What does this do" could mean the current file, the last Copilot response or a specific code block. Be specific instead, for example "What does the createUser function do?" or "What does the code in your last response do?"

• ASwitch to a premium model, which knows every library

  Incorrect

  No model knows every library; the guidance is to give context.

• BDescribe the library, or put its imports at the top of the file

  Correct

  If you use an uncommon library, describe it; if you want a specific library, set the import statements or say which library to use.

• CBlock suggestions matching public code so that the popular parser can't be used

  Incorrect

  The public code filter doesn't steer library choice.

• DDelete the popular parser from node_modules so Copilot can't see it

  Incorrect

  Copilot's suggestions aren't driven by installed packages in this way; context is.

Ambiguity also applies to libraries. If you are using an uncommon library, describe what it does. If you want to use a specific library, set the import statements at the top of the file or specify which library to use.

• A#project

  Incorrect

  #project includes the project context.

• B#file

  Incorrect

  #file includes the current file's content.

• C@terminal

  Incorrect

  @ introduces a chat participant; @terminal knows about the terminal, not your selection.

• D#selection

  Correct

  #selection includes the currently selected text in the prompt.

Chat variables include specific context in a prompt; type # followed by the variable. In VS Code, #selection includes the selected text, #file the current file, #function the current function, #block the current block and #project the project context. Chat participants are introduced with @.