NCP-GENL Prompt Engineering Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Zero-shot prompting with a detailed description of the desired function and its parameters.

  Why it's wrong here

  Zero-shot prompting relies solely on the model's pre-trained knowledge. If the model frequently uses deprecated libraries or incorrect signatures, a detailed description alone may not correct these issues because the model may not have up-to-date knowledge. Without examples, the model has no concrete pattern to follow for the specific libraries or signatures required, so it is likely to continue making the same mistakes.

• ✓

  Few-shot prompting with examples that demonstrate the correct use of the desired libraries and function signatures.

  Why this is correct

  Few-shot prompting provides the model with concrete examples of the desired output, including correct library usage and function signatures. By showing several input-output pairs where the output uses the correct libraries and signatures, the model can learn the pattern and apply it to new inputs. This is especially effective for code generation where precise syntax and API usage matter, and it directly addresses the deprecated library and signature issues.

• ✗

  Self-consistency prompting where the model generates multiple code solutions and the most common one is selected.

  Why it's wrong here

  Self-consistency involves generating multiple reasoning paths and selecting the most consistent answer. For code generation, if the model consistently uses a deprecated library, the most common solution will still be wrong. This technique is better for tasks with a single correct answer that can be voted on, but it does not correct systematic errors like using outdated APIs unless the majority of samples happen to be correct, which is unlikely if the model is biased.

• ✗

  Chain-of-thought prompting that asks the model to first explain its reasoning about which libraries to use, then write the code.

  Why it's wrong here

  Chain-of-thought can improve reasoning, but it does not guarantee correct library usage or signatures. The model might reason incorrectly about which libraries are current or how to call functions. Without examples of correct code, the model may still generate deprecated or incorrect code. While reasoning can help, it is not as direct as providing examples of the desired output for this specific problem.