Generative AI Leader Techniques to Improve Generative AI Model Output Practice Test 9 — 15 Questions

Question 1

A company is deploying a generative AI system that generates customer-facing emails. The system must ensure outputs are not toxic, biased, or harmful. Which TWO techniques are most effective for reducing toxicity in model outputs without significantly affecting performance?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Fine-tune the model on a dataset of safe emails using reinforcement learning from human feedback (RLHF).. Option C is correct because fine-tuning with RLHF directly aligns the model's behavior with human preferences for safe, non-toxic content, teaching it to avoid biased or harmful outputs at the model level without degrading overall generation quality. Option D is correct because a toxicity detection and filtering layer such as Vertex AI Safety Filters evaluates generated text against configurable harm categories and blocks or suppresses toxic outputs before they reach customers, adding a reliable guardrail. Option A is not effective because increasing max output tokens only changes length limits and can actually introduce more opportunities for harmful content. Option B is not effective because a very low temperature reduces randomness but does not remove learned toxic patterns and can hurt creativity and diversity. Option E is not effective because 50 few-shot examples consume significant context, increase cost and latency, and provide only shallow, prompt-level guidance that is easily bypassed.

Answer

Increase the maximum output token count to allow more context.

Answer

Set temperature to a very low value (e.g., 0.1).

Answer

Provide 50 few-shot examples of safe emails in every prompt.

Question 2

A company uses a text generation model for customer support but notices it occasionally provides outdated information. Which technique should they implement to improve output accuracy?

Accepted Answer

Implement retrieval-augmented generation (RAG). Retrieval-augmented generation (RAG) is the correct technique because it grounds the model's output in real-time, external knowledge sources (e.g., a vector database or document index) rather than relying solely on static training data. This directly addresses the problem of outdated information by allowing the model to retrieve and synthesize current facts at inference time, ensuring accuracy without requiring retraining.

Answer

Increase max output tokens

Answer

Fine-tune the model with more historical support data

Answer

Increase model temperature to 1.0

Question 3

Which TWO techniques are most effective for improving the quality of a generative AI model's output when summarizing complex documents?

Accepted Answer

Providing few-shot examples of ideal summaries. Option A is correct because few-shot prompting supplies the model with concrete examples of the desired summary format, style, and level of detail, which conditions the model to reproduce that structure and improves fidelity when summarizing complex documents. Option B is correct because a larger, more capable model such as PaLM 2 has greater capacity to comprehend long, intricate documents and generate more accurate, coherent summaries than a smaller predecessor like PaLM. Option C is not appropriate because increasing max output length only allows longer responses; it does not improve summary quality and can even encourage verbosity rather than conciseness. Option D is not appropriate because setting top_p to 0.1 sharply narrows nucleus sampling, reducing diversity and making the output more rigid without addressing summarization quality. Option E is not appropriate because raising temperature to 0.8 increases randomness, which tends to introduce inaccuracies and inconsistency rather than improve the quality of document summaries.

Answer

Increasing max output length significantly

Answer

Setting top_p to 0.1

Answer

Adjusting temperature to 0.8

Question 4

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?

Accepted Answer

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

Lower temperature to 0.1

Answer

Increase topP to 1.0

Answer

Increase maxOutputTokens to 1024

Question 5

A financial technology company has deployed a custom-tuned PaLM 2 model on Vertex AI to generate personalized investment recommendations for retail clients. The model was fine-tuned on a corpus of historical market data and advisory transcripts. Recently, the compliance team flagged that several recommendations contradicted SEC guidelines, and the model sometimes repeated prohibited statements from outdated training materials. The team has already implemented safety filters (e.g., blocking toxic content) and adjusted the model's system instructions to be more conservative. However, the issues persist. The model's deployment parameters are: temperature=0.4, top_p=0.9, max_output_tokens=500, and no grounding. The company must maintain compliance without significantly increasing latency. What should they do next?

Accepted Answer

Configure Vertex AI grounding using a curated data store of real-time SEC regulations and market data. Configuring Vertex AI grounding with a curated data store of real-time SEC regulations directly addresses the root cause: the model is generating outputs that contradict current compliance rules. Grounding forces the model to base its responses on authoritative, up-to-date sources, which is more effective than safety filters or system instructions alone, and it avoids the latency increase of a second model or the risk of catastrophic forgetting from additional fine-tuning.

Answer

Increase temperature to 0.7 to allow more diverse responses, and add a second model to verify outputs

Answer

Perform an additional fine-tuning round exclusively on the most recent SEC regulatory filings and compliance-approved content

Answer

Implement a chain-of-thought prompting technique that requires the model to explain its reasoning step by step

Question 6

A product team at a retailer is using Vertex AI Studio to build a Gemini-powered assistant that answers questions about their internal return policy. Early tests show the model invents policy details such as a 45-day return window, even though the official policy allows only 30 days. The team wants the assistant to answer strictly from a set of approved policy PDFs stored in a Cloud Storage bucket, and they want to avoid retraining the model. Which technique should they use?

Accepted Answer

Retrieval-augmented generation (RAG) with a Vertex AI Search data store grounded on the approved policy documents.. Grounding with retrieval-augmented generation lets the assistant fetch the relevant passages from the approved policy PDFs and condition its answer on that retrieved text, so the 30-day window is stated correctly and the response can cite the source. It requires no retraining, updates automatically when documents change, and directly addresses hallucinated policy details.

Answer

Adding a system instruction telling the model to be accurate and to never hallucinate policy details.

Answer

Supervised fine-tuning of the Gemini model on a labeled dataset of correct policy answers.

Answer

Lowering the model's temperature to 0 and increasing the top-k value in the generation configuration.

Question 7

A marketing team uses a generative AI model on Vertex AI to create ad copy variations. The outputs are often too random and include off-brand phrases. They want to reduce randomness while still allowing some creativity. Which parameter should they adjust?

Accepted Answer

Temperature. Temperature is the primary parameter that scales the randomness of token selection. Lowering temperature makes the model more likely to choose high-probability tokens, resulting in more predictable and brand-safe output. While top-K and top-P also influence diversity, they are secondary to temperature for controlling overall randomness.

Answer

Top-P

Answer

Top-K

Answer

Max output tokens

Question 8

A company is using a generative AI model to answer customer questions. The answers are often vague and do not use the company's specific terminology. They want to improve the relevance and specificity of the responses. Which technique should they use?

Accepted Answer

Using retrieval-augmented generation (RAG). Retrieval-augmented generation (RAG) enhances a generative model by retrieving relevant information from a knowledge base and including it in the prompt. This grounds the model's responses in factual, company-specific content, improving relevance and ensuring the use of correct terminology. It is efficient because it does not require retraining the model.

Answer

Increasing the model's temperature

Answer

Fine-tuning the model on company documents

Answer

Reducing the max output tokens

Question 9

A media company uses Gemini on Vertex AI to generate short news summaries from long articles. The summaries frequently miss key facts and sometimes include details not in the source. The team wants to improve factual grounding and coverage without retraining the model. (Choose two.)

Accepted Answer

Use retrieval-augmented generation (RAG) with Vertex AI Search to supply relevant source passages in the prompt.. Retrieval-augmented generation supplies authoritative source passages at inference time, and few-shot examples teach the model the desired grounded summarization pattern. Together they improve fact coverage and reduce unsupported claims without retraining. Temperature tuning, larger machines, and streaming do not add source knowledge or enforce faithfulness, so they fail to solve the stated problem.

Answer

Enable streaming responses so the model can revise earlier sentences as it generates.

Answer

Deploy the model on a larger machine type with more vCPUs to increase factual accuracy.

Answer

Increase the model temperature to 1.5 so the model explores more of the source content.

Question 10

A product team uses Gemini via the Vertex AI API to draft customer emails. The drafts are accurate but often too long and include unnecessary background. The team wants shorter, more direct outputs while keeping the same model. Which approach should they take?

Accepted Answer

Revise the prompt to explicitly instruct concise, direct language and specify a target length or format.. The most direct and efficient fix for verbose outputs is to change the prompt to request concise, direct language with a specified length or format. Prompt engineering shapes style immediately without retraining or infrastructure changes. Token limits truncate rather than summarize, fine-tuning is overkill for a style tweak, and top-p affects randomness rather than length.

Answer

Fine-tune the model on a dataset of short emails to permanently change its verbosity.

Answer

Set the max output tokens parameter to a very low value and leave the prompt unchanged.

Answer

Increase the top-p value so the model selects from a narrower set of likely tokens.

Question 11

A retail bank uses a Gemini model on Vertex AI to answer customer questions about its mortgage products. Testers report that the model sometimes invents interest rates that do not exist in the bank's rate sheet. The team wants the model to ground its answers only in an approved corpus of product documents and cite the passages it used. Which approach should they implement?

Accepted Answer

Configure Vertex AI Search grounding with the bank's document store and enable citations in the generateContent request.. Grounding with Vertex AI Search connects the model to the bank's authoritative document store, so answers are conditioned on retrieved passages rather than on parametric memory, and the grounding metadata enables citations. Sampling controls only shape randomness, fine-tuning bakes in a snapshot of knowledge without provenance, and token limits only change length.

Answer

Lower the temperature parameter to 0 and increase the topK value so the model samples only from the most probable tokens.

Answer

Fine-tune the Gemini model on the bank's historical customer support transcripts so it learns the correct rates.

Answer

Increase the model's output token limit so it has more room to state the correct rate before answering.

Question 12

A financial services firm uses a generative AI model on Vertex AI to answer employees' HR policy questions. The model sometimes invents policy details. The firm wants answers grounded in the official HR handbook and needs to cite the source section. Which solution should they implement?

Accepted Answer

Use retrieval-augmented generation with Vertex AI Search over the HR handbook and instruct the model to cite retrieved sections.. Retrieval-augmented generation over the HR handbook supplies authoritative passages and enables citations, keeping answers grounded and current. Fine-tuning embeds knowledge but does not guarantee citation or easy updates. Low temperature does not add missing knowledge, and stuffing the full handbook into the prompt is inefficient and less precise than retrieval.

Answer

Fine-tune the model on the HR handbook text so the knowledge is embedded in the weights.

Answer

Increase the context window by using a model with a larger token limit and paste the entire handbook into every prompt.

Answer

Lower the temperature to zero and rely on the model's pretrained knowledge of HR policies.

Question 13

A marketing team uses a Gemini model through the Vertex AI API to draft campaign copy. The drafts are creative but frequently wander off topic and include unsupported claims. The team wants a low-effort improvement before considering any model customization. Which action should they take first?

Accepted Answer

Rewrite the prompt to include a clear role, the target audience, explicit constraints, and a required output format.. Prompt engineering with an explicit role, audience, constraints, and output format is the fastest and cheapest way to align model output with a brief. Fine-tuning, larger models, and higher temperature all add cost or increase variance without addressing the underlying ambiguity in the instructions.

Answer

Raise the temperature so the model explores more ideas and eventually lands on better copy.

Answer

Switch the model to a larger version with more parameters to improve instruction following.

Answer

Create a supervised fine-tuning job with a few hundred labeled campaign examples.

Question 14

A marketing team is using a generative AI model on Vertex AI to create ad copy for a new product launch. The initial outputs are generic and do not reflect the brand's tone. The team wants to quickly improve the outputs without retraining the model. They have a set of example ad copies that exemplify the desired tone. Which technique should they use?

Accepted Answer

Use few-shot prompting by including the example ad copies in the prompt.. Few-shot prompting is ideal when you have example outputs that demonstrate the desired style or format. By including these examples in the prompt, the model can infer the pattern and generate new content that aligns with the brand's tone. This approach requires no training and can be implemented immediately, making it the most efficient solution for the marketing team.

Answer

Increase the model's temperature setting to encourage more creative outputs.

Answer

Deploy the model to a new endpoint with higher throughput.

Answer

Fine-tune the model on the example ad copies.

Question 15

A healthcare analytics team uses Gemini on Vertex AI to extract structured data from clinical notes. The model occasionally outputs invalid JSON, breaking downstream processing. The team wants to enforce a strict output schema. Which approach should they use?

Accepted Answer

Use controlled generation with a response schema in the Gemini API request to constrain output to the defined JSON structure.. Controlled generation with a response schema constrains decoding so the model emits JSON that conforms to the defined structure, guaranteeing parseable output. Prompt hints and low temperature reduce but do not eliminate format errors, fine-tuning is heavier and still not absolute, and regex repair is fragile. Schema-constrained decoding is the correct tool for strict structured output.

Answer

Fine-tune the model on a dataset of clinical notes paired with valid JSON outputs.

Answer

Post-process the model output with a regular expression to fix malformed JSON before parsing.

Answer

Add 'return JSON' to the prompt and set temperature to 0.2 to encourage valid formatting.