AIF-C01 ยท Responsible and explainable AI
Features of responsible AI
Bias, fairness, inclusivity, robustness, safety and veracity, and Amazon Bedrock Guardrails as a tool to enforce them.
Key points
- Fairness and low bias: similar quality of outcomes for different groups, plus inclusivity.
- Robustness: stable behaviour on unusual or noisy input.
- Safety: no harmful output.
- Veracity: truthful, grounded answers.
- Amazon Bedrock Guardrails: helps enforce safety and veracity at run time with content filters, denied topics, word filters, sensitive information filters and contextual grounding checks.
Source: https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails.html
Practice questions
Which Amazon Bedrock Guardrails safeguard can block or mask personally identifiable information in prompts and responses?
- Word filters
- Sensitive information filters
- Contextual grounding checks
- Denied topics
Show answer
Answer: B. Sensitive information filters
Sensitive information filters detect PII entities and custom regex patterns. Word filters only match fixed words or phrases you list.
Which of these are safeguards offered by Amazon Bedrock Guardrails?
- Content filters for categories such as hate or violence
- Automatic GPU autoscaling for inference
- Word filters for custom blocked phrases
- Contextual grounding checks for hallucinations
- Scheduled model retraining
Show answer
Answer: A. Content filters for categories such as hate or violence, C. Word filters for custom blocked phrases, D. Contextual grounding checks for hallucinations
Content filters, word filters and contextual grounding checks are guardrail policies. Autoscaling GPUs and scheduling retraining are infrastructure and MLOps tasks, not guardrail policies.
True or false: Amazon Bedrock Guardrails only inspects model responses, never the user's prompt.
Show answer
Answer: False
A guardrail evaluates both the input prompt and the model completion against the configured filters.
Practice 8 questions on this topic in Bitzu
Short lessons, instant explanations and spaced review of what you get wrong. Free on iOS and Android.
