Generative AI Leader Practice Question: A developer is using Vertex AI Studio to…

Answer analysis

Option-by-option breakdown

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

• ✗

  Safety filters

  Why it's wrong here

  Safety filters block harmful or policy-violating content categories; they do not retrieve or cite trusted source material, so grounding is unaffected. It is tempting because filters are a familiar Vertex AI Studio control, and enabling them would be correct when the concern is toxic, violent or otherwise disallowed output rather than factual accuracy.

• ✗

  Temperature setting reduction

  Why it's wrong here

  Temperature reduction narrows sampling randomness, affecting phrasing and variability, but it does not connect the model to any external corpus, so factual grounding cannot result. It is tempting because lower temperature is associated with less hallucination, and it would be correct when the model already has the needed facts but expresses them inconsistently.

• ✗

  Chain-of-thought prompting

  Why it's wrong here

  Chain-of-thought prompting elicits intermediate reasoning steps within the model's own parameters; it supplies no external evidence, so responses remain ungrounded. It is tempting because it improves multi-step accuracy, and it would be correct for arithmetic, logic or planning tasks where the model already holds the required knowledge.

• ✓

  Grounding with a Vertex AI Search data store

  Why this is correct

  Grounding with a Vertex AI Search data store retrieves passages from the trusted knowledge base and supplies them to the model, so responses cite verifiable sources rather than relying on parametric memory. This satisfies the requirement that answers be grounded in factual information.

Go deeper

Related to this question