Techniques to Improve Generative AI Model Output →hardMultiple Choice

Generative AI Leader Practice Question: Techniques to Improve Generative AI Model Output

Exhibit

Refer to the exhibit.

```json
{
  "model": "publishers/google/models/chat-bison@001",
  "endpoint": "us-central1-aiplatform.googleapis.com",
  "parameters": {
    "temperature": 0.9,
    "topP": 0.95,
    "maxOutputTokens": 256,
    "groundingConfig": {
      "sources": []
    }
  },
  "deployment": "production"
}
```

The exhibit shows the deployment configuration for a conversational AI model used in a finance application. Users report that responses are creative but often contain factually incorrect financial advice. Which parameter change would most improve factual accuracy?

⚠ Common exam trap

The Generative AI Leader exam often tests the misconception that adjusting sampling parameters (temperature, topP) can fix factual accuracy issues, when in reality those parameters only control output randomness and diversity, not the truthfulness of the underlying knowledge.

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 grounding sources, such as "EnterpriseSearch" or "Web"

Adding grounding sources like EnterpriseSearch or Web provides the model with access to authoritative, up-to-date financial data, which directly reduces hallucinations by anchoring responses in verified facts rather than relying solely on the model's parametric knowledge. This is the most effective technique for improving factual accuracy in a domain where correctness is critical.

Answer analysis

Option-by-option breakdown

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

✓
Add grounding sources, such as "EnterpriseSearch" or "Web"
Why this is correct
Adding grounding sources constrains generation to retrieved enterprise or web content, so the model cites evidence rather than relying on parametric memory. This directly addresses the stem's factual-accuracy failure, where ungrounded creativity produces incorrect financial advice. Retrieval augmentation, not sampling parameters, is the mechanism that anchors responses in verifiable data.
✗
Lower temperature to 0.1
Why it's wrong here
Lowering temperature to 0.1 reduces sampling randomness, making token selection deterministic and factually stable. It is tempting as the standard creativity dial, but it cannot inject missing financial knowledge; retrieval grounding or fine-tuning addresses factual accuracy, not decoding parameters.
✗
Increase topP to 1.0
Why it's wrong here
Raising topP to 1.0 widens nucleus sampling to the full probability mass, increasing randomness and factual drift. It is tempting because higher values appear to give the model more choice, and would be correct when generating varied creative prose rather than regulated financial guidance.
✗
Increase maxOutputTokens to 1024
Why it's wrong here
maxOutputTokens caps response length only; it does not alter the sampling distribution that produces fabricated financial claims. It is tempting because longer answers feel more thorough, and would be correct when replies are being truncated mid-sentence before the required detail is delivered.

Go deeper

Generative AI Leader Practice Question: Techniques to Improve Generative AI Model Output

Exhibit

Why each option matters

Option-by-option breakdown

Related to this question