GCP Generative AI Leader Practice Question 78: Safety Settings
Google Cloud Generative AI Leader · Test 8 · question 8 of 10
A media company is building a public-facing gen AI assistant and wants to configure how strictly the model should block responses that its content classifier scores as likely to contain harassment, hate speech, or sexually explicit material, choosing a stricter or looser threshold for each harm category.
Which model configuration option is designed for this?
Show answer and explanation
Answer:
Safety settings are a sampling-time configuration that let a developer set how likely content in each harm category needs to be before the response is blocked, covering categories such as harassment, hate speech, dangerous content, and sexually explicit material. That is exactly the control the media company needs to tune.
A is wrong because top-p (nucleus sampling) controls the diversity of token choices during generation, not content safety filtering. C is wrong because output length limits how long a response can be, unrelated to blocking harmful content categories. D is wrong because temperature controls the randomness of word choice, not whether a response is screened for harmful content.
Reference: Safety settings