Generative AI Leader Practice Question: Techniques to Improve Generative AI Model Output
After fine-tuning a model on customer support data, the model starts using profanity. What is the most effective mitigation?
⚠ Common exam trap
Google often tests the misconception that modifying training parameters (like learning rate or temperature) can fix output quality issues, when in fact post-processing filters are the standard, immediate solution for content safety in production LLM systems.
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
✓
Enable a safety attribute filter
Enabling a safety attribute filter is the most effective mitigation because it acts as a post-processing guardrail that blocks profanity at inference time, regardless of the model's training data. This is a standard practice in production LLM deployments, where safety filters (e.g., using keyword matching or classifier models) intercept and redact harmful outputs before they reach the user, providing immediate and reliable control without requiring retraining.
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 profanity to training data as negative examples
Why it's wrong here
Labelling profanity as negative examples still exposes the model to those tokens and typically teaches it to reproduce them in other contexts. It is tempting because supervised fine-tuning with labelled examples is the standard way to steer model behaviour, and negative examples work for classification tasks rather than generative text.
- ✗
Reduce learning rate and retrain
Why it's wrong here
Lowering the learning rate only slows weight updates; it does not remove the profane patterns already present in the fine-tuning corpus, so the behaviour persists. It is tempting because learning rate governs how aggressively the model adapts, and tuning it is a standard remedy when training is unstable or overfitting.
- ✗
Increase temperature to reduce confidence
Why it's wrong here
Temperature controls sampling randomness, not the learned probability distribution; raising it makes profane tokens more likely to be sampled, not less. It is tempting because temperature is the usual knob for controlling output variability, and lowering it would make existing bad tendencies more deterministic rather than removing them.
- ✓
Enable a safety attribute filter
Why this is correct
A safety attribute filter intercepts model outputs and blocks or redacts harmful content such as profanity before it reaches users, directly addressing the fine-tuning side effect. Retraining or prompt engineering may reduce but not reliably prevent toxic generations.
Go deeper
Related to this question
About these practice questions
One of 1,008 original Generative AI Leader practice questions on Courseiva, each with a full explanation and wrong-answer analysis — not exam dumps or protected exam content. Learn why practice questions differ from exam dumps →
JA
Written by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
This Generative AI Leader practice question is part of Courseiva's free Google Cloud certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the Generative AI Leader exam.