GCP Generative AI Leader Practice Question 57: Prompt Chaining Subtasks
Google Cloud Generative AI Leader · Test 6 · question 7 of 10
A single prompt asks a model to read a long contract, extract every payment obligation, classify each one by risk level, and then draft a plain-language summary for a non-lawyer. In practice the model often skips obligations, mislabels risk, or blends the extraction and the summary together inconsistently. The team wants more reliable, easier-to-debug output without changing the model.
Which prompting technique addresses this best?
Show answer and explanation
Answer:
Prompt chaining breaks a complex task into a sequence of simpler prompts, where the output of one becomes the input to the next. Here, that means a dedicated extraction prompt produces a list of payment obligations, a second prompt classifies each one by risk, and a third prompt turns the classified list into a plain-language summary. Each step is narrower and easier to get right, and each intermediate output can be inspected, which improves both accuracy and debuggability compared to asking one prompt to do everything at once.
- A is wrong because raising temperature increases randomness in wording, not the model's ability to reliably follow multiple sequential instructions in one pass; it can make output less consistent. - B is wrong because more examples of the same all-in-one task still ask the model to juggle extraction, classification, and summarization simultaneously; it does not reduce the task's complexity. - D is wrong because sampling multiple full attempts and picking one manually does not fix the underlying problem of an overloaded single prompt, and it does not scale as a repeatable process.