Recommend an Alternative for Problematic Text Mitigation in a Data Source Feeding a GenAI Application GENERATIVE-AI-ENGINEER-ASSOCIATE Practice Questions (Page 2)

Questions 6–10

1. 6foundation · easy

   A generative AI application uses a public dataset that contains offensive language. The governance team requires that no offensive content appears in the model's training data. The dataset is large, and the offensive content is scattered across many records. Which mitigation approach is most suitable?

   • ARedacting the offensive words
   • BExcluding the entire dataset
   • CRewriting the offensive content
   • DFiltering out all records containing offensive language
   Reveal Solution Discussion Select an answer first
2. 7expert · hard

   A telecommunications company is building a customer-service chatbot that uses a knowledge base of support tickets. The tickets contain customer phone numbers and account details. The company wants to prevent the chatbot from exposing this information, but the tickets are the only source of troubleshooting information. The company also wants to retain the ability to escalate a ticket to a human agent when needed. Which mitigation strategy is most appropriate?

   • AExclude all tickets that contain phone numbers or account details.
   • BUse a prompt that instructs the model to never reveal phone numbers or account details.
   • CAdd a response filter that blocks any output containing a phone number or account number.
   • DRedact the phone numbers and account details from the tickets before indexing them for retrieval, and maintain a separate mapping table for escalation.
   Reveal Solution Discussion Select an answer first
4. 8foundation · easy

   A dataset contains customer reviews that include personal information such as phone numbers. The team wants to keep the reviews for sentiment analysis but must protect the individuals' privacy. Which mitigation strategy is most appropriate?

   • AExcluding the entire dataset
   • BRedacting the personal information
   • CRewriting the reviews to remove personal details
   • DFiltering out all reviews that contain any personal information
   Reveal Solution Discussion Select an answer first
5. 9expert · hard

   A government agency is building a chatbot to answer citizen questions about public services. The chatbot uses a knowledge base of internal documents that contain both public information and sensitive internal notes (e.g., employee names, internal procedures). The agency wants to make the chatbot publicly accessible, but must not expose the sensitive information. The sensitive information is scattered throughout the documents. Which mitigation strategy is most appropriate?

   • AUse a prompt that instructs the model to only answer questions about public services.
   • BExclude the entire documents that contain any sensitive information.
   • CAdd a response filter that blocks any output containing employee names or internal procedures.
   • DRedact the sensitive information from the documents before indexing them for retrieval.
   Reveal Solution Discussion Select an answer first
6. 10application · medium

   A news organization is building a chatbot to summarize articles for readers. The articles contain quotes from public figures, some of which include offensive language. The organization wants the chatbot to provide neutral and professional summaries. The offensive quotes are a small part of the articles. Which mitigation strategy is most appropriate?

   • ARewrite the offensive quotes to be non-offensive while preserving the meaning.
   • BRedact the offensive words and replace them with a placeholder.
   • CFilter out the entire articles that contain offensive quotes.
   • DExclude all articles from authors who have ever used offensive language.
   Reveal Solution Discussion Select an answer first