Implement AI Solutions Responsibly AI-102 Practice Questions (Page 6)

Questions 26–30

1. 26application · medium

   A social media platform uses Azure AI Content Safety to moderate user-generated images. The moderation team reports that while the service correctly flags violent content, it is missing many posts containing hate symbols that are embedded in memes. The company wants to reduce the number of missed items without blocking all meme posts. What should you do?

   • AAdd the specific hate symbols to a custom blocklist and configure the moderation workflow to check images against that blocklist before applying the standard severity thresholds.
   • BSwitch the moderation service to Azure Computer Vision and use optical character recognition to extract all text from memes, then run a keyword search for hate symbols.
   • CIncrease the severity threshold for the violence category so that more images are flagged, including those with hate symbols.
   • DDisable the violence category and rely solely on the hate and fairness category, which is designed to detect hate symbols in images.
   Reveal Solution Discussion Select an answer first
2. 27foundation · easy

   Which Azure AI service is specifically designed to help detect and filter potentially offensive or inappropriate content in text and images?

   • AAzure AI Translator
   • BAzure AI Language
   • CAzure AI Vision
   • DAzure AI Content Safety
   Reveal Solution Discussion Select an answer first
4. 28expert · hard

   A social media company is deploying an Azure OpenAI-based content summarization tool for news articles. They are concerned about both prompt injection attacks (where users try to make the tool ignore its instructions) and the tool generating summaries that contain hate speech. They also want to block a specific set of known hate symbols from appearing in the summaries. What should they implement?

   • AUse groundedness detection with a database of approved news sources, and enable prompt shields.
   • BAdd all hate symbols to a custom blocklist and set the action to 'Reject'.
   • CEnable prompt shields, configure content filters for hate speech, and add the hate symbols to a custom blocklist.
   • DConfigure content filters with a high severity threshold and rely on the model's safety training.
   Reveal Solution Discussion Select an answer first
5. 29application · medium

   A news website uses Azure AI Content Safety to moderate comments. The site wants to block hate speech and also prevent users from using alternative spellings to bypass the filter. What should you configure?

   • AUse only the self-harm content filter, since it is the most comprehensive.
   • BDisable all content filters and rely on user reports to identify hate speech.
   • CSet the severity threshold to the lowest level to flag all comments for manual review.
   • DUse the hate and fairness content filter and enable a custom blocklist with variation matching for known hate terms.
   Reveal Solution Discussion Select an answer first
6. 30foundation · easy

   What is the purpose of a 'prompt shield' in an AI solution?

   • ATo encrypt the prompts sent to the AI model
   • BTo translate prompts into different languages
   • CTo detect and mitigate prompt injection attacks
   • DTo speed up the processing of prompts
   Reveal Solution Discussion Select an answer first