AIF-C01 Question 167
Select 2A data scientist is developing a generative AI application using Amazon Bedrock. To ensure the application adheres to responsible AI principles, they want to implement guardrails to control model responses and prevent harmful or biased outputs. Which of the following actions can they take to achieve this using Bedrock’s features?
- A
Use content filtering to block inappropriate or toxic language in model outputs.
- B
Enable model explainability to understand the reasoning behind each model prediction.
- C
Incorporate prompt templates to guide the model toward specific and safe responses.
- D
Leverage managed endpoint monitoring to track the usage of the underlying model.
- E
Set up sentiment analysis to identify and block negative sentiment in user inputs.
Show answer and explanation
Correct answers: A, C
Explanation
Amazon Bedrock provides features like content filtering and prompt templates to help developers implement responsible AI guardrails. Content filtering ensures the model avoids generating harmful or inappropriate responses, while prompt templates help guide the model to produce safe and contextually appropriate outputs. These features play a direct role in adhering to responsible AI principles.
- A. Correct.
Content filtering is a key feature of responsible AI guardrails in Amazon Bedrock, allowing you to block inappropriate or harmful outputs from the model.
- B. Incorrect.
While model explainability is important for understanding predictions, it is not directly related to implementing guardrails for responsible AI in Amazon Bedrock.
- C. Correct.
Prompt templates can be used to guide models toward generating safe and specific responses, aligning with responsible AI principles.
- D. Incorrect.
Managed endpoint monitoring is useful for tracking model usage and performance but does not directly address guardrails for responsible AI.
- E. Incorrect.
Sentiment analysis can detect the tone of user inputs but is not a specific guardrail feature provided by Amazon Bedrock.