Google Cloud's Generative AI Offerings →hardMultiple Choice

Generative AI Leader System Instructions Practice Question

Exhibit

Refer to the exhibit.

```
{
  "predictions": [
    {
      "content": "The capital of France is Paris.",
      "safetyAttributes": {
        "categories": ["Toxic"],
        "scores": [0.02],
        "blocked": false
      }
    }
  ],
  "metadata": {
    "tokenCount": 8
  }
}
```

A developer receives the above JSON response from a Vertex AI language model. The output content is correct, but the developer expected the model to not answer geography questions. What should the developer do to prevent the model from responding to geography queries?

⚠ Common exam trap

Candidates often assume safety filters can be configured for arbitrary topics like geography, but Vertex AI safety filters only support predefined categories. The correct approach is to use system instructions to guide model behavior for custom restrictions.

Answer choices

Why each option matters

Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.

Correct answer & explanation

✓

Add a system instruction to not answer geography questions

Vertex AI does not have a predefined safety filter category for 'Geography'. To prevent the model from answering geography questions, the developer should use a system instruction that explicitly tells the model not to respond to geography queries. System instructions are the appropriate mechanism for custom content restrictions, while safety filters are limited to predefined categories such as toxic, harassment, etc.

Answer analysis

Option-by-option breakdown

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

✗
Adjust the safety filter thresholds for the 'Toxic' category
Why it's wrong here
Safety filters govern harmful-content categories such as toxicity, harassment and violence; geography is a topical restriction, not a safety harm, so threshold changes never block it. It tempts because safety filters are the standard control for refusing unwanted model output, and would be correct if the queries sought dangerous or abusive content.
✗
Enable Vertex AI Grounding with a geography knowledge base
Why it's wrong here
Grounding supplies authoritative external data to improve factual accuracy; it does not restrict which topics the model may answer, so geography replies continue. It is tempting because grounding is the usual fix for incorrect model output, and that would be correct when answers need verifiable, up-to-date sources.
✗
Configure a safety filter for the 'Geography' category
Why it's wrong here
Vertex AI safety filters cover harm categories such as dangerous content, harassment and hate speech; geography is not one, so no such filter exists to configure. It is tempting because safety filters are the standard control for blocking unwanted model responses, and that would be correct for genuinely harmful content categories.
✓
Add a system instruction to not answer geography questions
Why this is correct
System instructions constrain model behaviour at inference time without retraining, directly satisfying the requirement to block geography responses. Unlike safety filters or fine-tuning, they apply per-request guidance that the model honours across turns, making them the appropriate control for steering a Vertex AI language model away from a chosen topic.

Go deeper

Generative AI Leader System Instructions Practice Question

Exhibit

Why each option matters

Option-by-option breakdown

Related to this question