Techniques to Improve Generative AI Model Output →hardMultiple Choice

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

A law firm uses a generative model to analyze contracts and extract key clauses. The model often outputs irrelevant clauses or misses important ones. They want to improve the relevance of the outputs without retraining the entire model. Which approach is best?

⚠ Common exam trap

A common trap in Google's Generative AI exams is assuming that adjusting parameters like temperature or token limits can fix output relevance issues. While these parameters affect randomness and context length, they do not guarantee that the model will generate accurate or relevant content for domain-specific tasks. The correct approach is to use a retrieval system like RAG to ground the model in a curated knowledge base.

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

✓

Implement Retrieval-Augmented Generation (RAG) with a curated legal clause database and a reranker to select the most on-topic passages.

Retrieval-Augmented Generation (RAG) with a curated legal clause database and a reranker is the best approach because it grounds the model's outputs in a trusted, external knowledge base, ensuring that only the most relevant clauses are retrieved and used for generation. This directly addresses the problem of irrelevant outputs and missed clauses without requiring retraining, as the model can dynamically fetch and rank the most on-topic passages from the curated database.

Answer analysis

Option-by-option breakdown

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

✗
Increase the input token limit to provide the entire contract in the prompt.
Why it's wrong here
A larger context window does not improve clause selection; irrelevant text still competes for attention and can dilute relevance. It is tempting because stuffing the whole contract avoids chunking, and that approach suits summarisation or whole-document question answering, not precise extraction of specific clauses.
✗
Decrease the temperature to make outputs more deterministic.
Why it's wrong here
Lower temperature reduces randomness in wording, not relevance of retrieved content; the model can still omit or invent clauses. It is tempting because determinism feels like accuracy, and it is the right lever for reproducible formatting or classification, but extraction relevance depends on grounding, not sampling.
✓
Implement Retrieval-Augmented Generation (RAG) with a curated legal clause database and a reranker to select the most on-topic passages.
Why this is correct
RAG retrieves passages from a curated legal clause database at inference time and a reranker scores them for topical relevance, grounding generation in authoritative clauses. This improves precision and recall of extracted clauses without retraining, satisfying the constraint of no full model retraining.
✗
Fine-tune the base model on a labeled dataset of contract-clause pairs.
Why it's wrong here
Fine-tuning changes model weights, which the scenario explicitly excludes by requiring improvement without retraining. It is tempting because supervised tuning on contract-clause pairs genuinely improves domain extraction, and it would be correct if labelled data and retraining budget were available.

Go deeper

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

Why each option matters

Option-by-option breakdown

Related to this question