GENERATIVE-AI-LEADER practice question 10 of 10

A development team is configuring a generative AI model for a customer-facing application and wants to ensure the generated content is appropriate and harmless. What is the primary function of the safety settings parameter in a generative AI model?

Full set: 76 questions

Domain Innovating with data and Google Cloud

Answer options
A: To limit the maximum text length that the model generates by ensuring concise responses.
B: To determine the number of tokens the model can process at once by influencing the complexity and length of inputs and outputs.
C: To filter out potentially harmful or inappropriate content from the model's output based on the desired level of filtering. Correct answer
D: To control the creativity and randomness of the model's output by adjusting the diversity of word choices.
Correct Answer:
C. To filter out potentially harmful or inappropriate content from the model's output based on the desired level of filtering.
Explanation

The primary function of safety settings in a generative AI model is to implement responsible AI practices by filtering potentially harmful content. These settings allow developers to adjust the thresholds for various safety categories, such as harassment, hate speech, sexually explicit material, and dangerous content. By configuring these filters, the model can block responses that are likely to be inappropriate or unsafe, which is a critical requirement for customer-facing applications. This ensures that the generated output aligns with safety policies and protects end-users from harmful interactions.

Why Incorrect

A. Limiting text length is controlled by parameters like maxoutputtokens, not safety settings.
B. The number of tokens a model can process relates to its context window, a fundamental model attribute, not a configurable safety filter.
D. Controlling creativity and randomness is the function of parameters such as temperature, top-k, and top-p.

References

1. Google Cloud Documentation

"Safety filter thresholds": "Use safety filters to adjust the likelihood of getting responses that may be harmful. By default

safety filters block content...with a medium or high probability of being unsafe across all safety dimensions." This document details the categories (Harassment

Hate speech

etc.) and block thresholds

directly supporting the function described in option C.

Source: Google Cloud

Vertex AI Documentation

"Gemini API"

"Safety settings".

2. Google Cloud Documentation

"Model parameters": This documentation explicitly separates parameters for controlling creativity (temperature

topK

topP) and response length (maxOutputTokens) from safety settings

confirming they serve different primary functions.

Source: Google Cloud

Vertex AI Documentation

"Generative AI on Vertex AI"

"Model parameters".

3. Google AI

"A new generation of models for safety": "We’ve also built dedicated safety classifiers to identify

label

and stop harmful content... These classifiers are part of a multi-layered safety strategy." This highlights that safety is a distinct

classified filtering function.

Source: Google AI

Responsible AI

"A new generation of models for safety" section.

Scroll to Top