Generative AI Leader Techniques to Improve Generative AI Model Output Practice Test 6 — 15 Questions

Question 1

A team wants to reduce hallucinations in a question-answering model. Which THREE techniques should they consider?

Accepted Answer

Fine-tune the model on a curated factual dataset. Fine-tuning on a curated factual dataset directly adjusts the model's weights to prioritize accurate, domain-specific knowledge, reducing the likelihood of generating unsupported or hallucinated content. This technique anchors the model's output in verified data, making it more reliable for question-answering tasks.

Answer

Reduce the number of tokens in output

Answer

Increase the temperature parameter

Question 2

Which TWO techniques are commonly used to control the style and tone of a generative model's output?

Accepted Answer

Fine-tuning on a dataset with desired style. Fine-tuning on a dataset that embodies the desired style directly adjusts the model's weights, making it consistently produce outputs with that specific tone and style. This is a fundamental technique for customizing generative models, as it teaches the model the exact patterns, vocabulary, and stylistic nuances present in the training data.

Answer

Adjusting the temperature

Answer

Modifying the top_k value

Answer

Changing the top_p value

Question 3

Refer to the exhibit. The endpoint is experiencing high latency during traffic spikes. The team wants to improve response time by reducing queueing. Which change to the configuration would be most effective?

Accepted Answer

Increase maxReplicaCount to 10. Increasing maxReplicaCount to 10 allows the autoscaler to provision more replicas during traffic spikes, distributing the incoming requests across additional endpoints. This directly reduces queueing at each replica because the load is spread over more instances, lowering per-instance latency. The change targets the root cause—insufficient capacity to handle peak load—rather than adjusting thresholds or model versions.

Answer

Decrease minReplicaCount to 0

Answer

Change the model version to '2'

Answer

Decrease the target value in autoscaling metric to 50

Question 4

A data science team is fine-tuning a large language model using Vertex AI to generate marketing copy. They notice that the generated text is often repetitive and lacks creativity. Which technique should they apply to improve output diversity?

Accepted Answer

Increase the temperature parameter to 0.9.. Increasing the temperature parameter to 0.9 raises the randomness of the probability distribution over tokens, allowing less likely tokens to be selected. This directly counteracts repetitive output by encouraging the model to explore more diverse word choices, which is a standard technique for improving creativity in text generation.

Answer

Decrease the beam search width to 1.

Answer

Decrease the top-k sampling threshold.

Answer

Add more examples of repetitive text to the training dataset.

Question 5

A team is using Vertex AI Pipelines to deploy a generative AI model for real-time inference. The model sometimes generates harmful content. They want to implement a safety filter that checks the output before returning it to the user, but they need to minimize latency. Which approach best balances safety and performance?

Accepted Answer

Use a secondary lightweight classifier to filter outputs in real-time.. Deploying a secondary lightweight classifier (e.g., a distilled BERT or a small logistic regression model) as a post-processing filter allows real-time inference with minimal latency overhead. This approach decouples safety from the primary generative model, enabling fast rejection of harmful outputs without retraining or blocking the main inference pipeline.

Answer

Retrain the model on every flagged harmful output.

Answer

Manually review all outputs before delivery.

Answer

Disable safety checks to improve latency.

Question 6

A developer is using the Gemini API to build a chatbot. They want the model to always respond in a friendly, professional tone. Which prompt engineering technique should they use?

Accepted Answer

Set system instructions to 'You are a friendly and professional assistant.'. Setting system instructions is the most direct and reliable way to define the model's persona and behavioral constraints. In the Gemini API, system instructions act as a persistent, top-level directive that influences every response, ensuring the chatbot consistently adopts a friendly and professional tone without requiring repeated examples or parameter tuning.

Answer

Include a few-shot example in every user message.

Answer

Set the temperature to 0.2.

Answer

Set max output tokens to 100.

Question 7

A team is using a pre-trained language model to summarize legal documents. They find that summaries often miss key dates and parties involved. Which technique would most effectively improve factual accuracy?

Accepted Answer

Fine-tune the model on a dataset of legal summaries with annotated key entities.. Fine-tuning on a dataset of legal summaries with annotated key entities directly teaches the model to recognize and reproduce critical factual elements like dates and parties. This supervised learning approach adjusts the model's weights to prioritize entity extraction and accurate generation, which is the most effective method for improving factual accuracy in domain-specific tasks.

Answer

Use top-p sampling with a low p value.

Answer

Increase the temperature parameter.

Answer

Use chain-of-thought prompting.

Question 8

A company is deploying a generative AI model for customer support. They want to reduce hallucinations while maintaining fluency. They have a large dataset of previous support conversations. Which strategy should they prioritize?

Accepted Answer

Implement retrieval-augmented generation (RAG) using the conversation dataset as a knowledge base.. Retrieval-augmented generation (RAG) directly addresses hallucinations by grounding the model's responses in factual, retrieved data from the conversation dataset. This approach allows the model to generate fluent, contextually relevant answers while reducing the risk of inventing information, as it retrieves actual support interactions as evidence before generating a response.

Answer

Increase the beam search width to 10.

Answer

Fine-tune the model on the conversation dataset.

Answer

Set the temperature to 0.1.

Question 9

A developer is using Vertex AI Studio to test prompts for a text generation model. They want the model to follow a specific output format (JSON). Which prompt engineering approach is most effective?

Accepted Answer

Include a few-shot example of the exact JSON format in the prompt.. Including a few-shot example of the exact JSON format in the prompt provides the model with a concrete pattern to follow, which is the most reliable method for enforcing structured output in generative models. Few-shot prompting leverages in-context learning, where the model uses the provided example to infer the desired schema and formatting rules, reducing ambiguity and improving adherence to the specified JSON structure.

Answer

Set stop sequences to '}'.

Answer

Set the system instruction to 'Always output JSON.'

Answer

Set temperature to 0 to make output deterministic.

Question 10

Which TWO techniques can help improve the factual accuracy of a language model's outputs? (Choose two.)

Accepted Answer

Fine-tune on a domain-specific curated dataset.. Fine-tuning on a domain-specific curated dataset (C) directly adjusts the model's weights using high-quality, verified examples, teaching it to produce factually correct outputs for that domain. This reduces hallucinations by grounding the model in accurate, relevant data rather than relying solely on its pre-training distribution.

Answer

Decrease the max output tokens.

Answer

Increase the temperature parameter.

Answer

Use top-k random sampling.

Question 11

Which THREE approaches are effective for reducing bias in generative model outputs? (Choose three.)

Accepted Answer

Use a balanced training dataset.. A balanced training dataset reduces the risk of the model learning spurious correlations or skewed distributions that lead to biased outputs. By ensuring that all demographic groups, topics, or perspectives are represented proportionally, the model's learned probability distribution is less likely to favor one group over another, directly mitigating representation bias at the data level.

Answer

Set temperature to a very high value.

Answer

Use adversarial training.

Question 12

A healthcare company is using a fine-tuned version of PaLM 2 on Vertex AI to generate clinical notes from doctor-patient conversations. The model was fine-tuned on a dataset of 10,000 de-identified transcripts and corresponding notes. During testing, the generated notes are grammatically correct and well-structured, but they often contain subtle inaccuracies: for example, they might mention a medication that was not discussed, or omit a key symptom. The team has already tried increasing the training epochs and adjusting learning rates, with minimal improvement. They need a solution that can be implemented quickly to improve factual accuracy without retraining the entire model. The team has access to a large archive of verified clinical notes and a small set of recent conversation-to-note pairs that have been manually reviewed and corrected. The inference pipeline currently uses a single call to the model with the conversation transcript as input. What should the team do?

Accepted Answer

Implement retrieval-augmented generation (RAG) by retrieving similar verified notes from the archive and providing them as context in the prompt.. Retrieval-augmented generation (RAG) directly addresses the core issue of factual inaccuracy without retraining. By retrieving verified clinical notes similar to the current conversation from the archive and injecting them as context in the prompt, the model gains access to ground-truth examples that anchor its output to factual details. This approach leverages the team's existing archive and small set of corrected pairs to provide relevant, accurate context, improving precision without modifying the model's weights.

Answer

Decrease the temperature to 0.1 to reduce randomness and force the model to stick to the input.

Answer

Use prompt engineering to instruct the model to only include information explicitly mentioned in the conversation.

Answer

Add a human-in-the-loop step to review and correct every generated note before use.

Question 13

You are a Generative AI architect at a large financial services firm. The firm has deployed a custom large language model (LLM) fine-tuned on proprietary financial reports to assist analysts in generating quarterly earnings summaries. The model is hosted on Vertex AI using a dedicated endpoint with autoscaling enabled. Recently, the model's output has exhibited two issues: (1) occasional factual inaccuracies about specific financial figures, and (2) a tendency to produce overly verbose and repetitive text in the summaries, sometimes exceeding the desired length of 200 words. The team has already tried adjusting the temperature parameter from 0.7 to 0.2 and increased the top-k sampling from 40 to 50, but the problems persist. The model's training data includes over 10,000 financial reports, and the fine-tuning process used low-rank adaptation (LoRA) with rank 16. The production environment uses a batch size of 1 for inference. You need to recommend a course of action that most directly addresses both the factual accuracy and verbosity issues without requiring a full retraining of the model. Which approach should you take?

Accepted Answer

Implement a retrieval-augmented generation (RAG) pipeline that queries a vector database of verified financial data, and apply constrained decoding with a maximum token limit and a repetition penalty.. It directly addresses both issues without retraining. A RAG pipeline grounds the model's outputs in verified financial data, eliminating factual inaccuracies. Constrained decoding with a maximum token limit and repetition penalty directly curbs verbosity and repetition, which temperature and top-k adjustments failed to fix.

Answer

Increase the LoRA rank to 32 and fine-tune the model for additional epochs on a curated subset of reports that focus on concise and accurate summaries.

Answer

Switch to a larger pre-trained model (e.g., PaLM 2 or GPT-4) and use the same fine-tuning data with higher rank LoRA to improve capability, then rely on the larger model's inherent accuracy.

Answer

Experiment with higher temperature (e.g., 0.9) and lower top-k (e.g., 20) to encourage more diverse and concise outputs, and add a post-processing step to truncate summaries to 200 words.

Question 14

A model generates responses that frequently repeat phrases or words. Which parameter adjustment is most likely to fix this?

Accepted Answer

Increase repetition penalty. Increasing the repetition penalty directly discourages the model from selecting tokens that have already appeared in the generated sequence, thereby reducing repetitive phrases or words. This parameter works by subtracting a fixed penalty from the logits of previously generated tokens before applying the softmax function, making them less likely to be chosen again.

Answer

Increase top_k

Answer

Increase temperature

Answer

Increase max output tokens

Question 15

To ensure that a generative AI model uses the most current information from the web for answering user queries, which Vertex AI feature should be enabled?

Accepted Answer

Grounding with Google Search. Grounding with Google Search is the correct feature because it enables the model to retrieve and reference real-time information from the web, ensuring responses are based on the most current data available. This is achieved by integrating Google Search results directly into the model's generation process, allowing it to cite live sources and reduce hallucinations from outdated training data.

Answer

Safety filters

Answer

Context caching

Answer

Model tuning