Recommend an Alternative for Problematic Text Mitigation in a Data Source Feeding a GenAI Application GENERATIVE-AI-ENGINEER-ASSOCIATE Practice Questions (Page 1)

Questions 1–5

1. 1application · medium

   A company is building a customer-support chatbot that uses a curated knowledge base of historical support tickets. During testing, the chatbot sometimes produces responses that include customer names and account numbers found in the tickets. The team needs to keep the tickets for context but must prevent the chatbot from exposing personal data. What should they do?

   • AExclude all tickets that contain any PII from the knowledge base.
   • BApply a PII redaction pipeline to the tickets before they are used for model training or retrieval.
   • CUse a filter that blocks any response containing a customer name or account number.
   • DAdd a system prompt instructing the chatbot to never output PII.
   Reveal Solution Discussion Select an answer first
2. 2foundation · easy

   A data engineer is reviewing a dataset that will be used to fine-tune a generative AI model. Which type of content in the dataset is most likely to cause the model to produce harmful or biased outputs?

   • AText that includes offensive language or hate speech
   • BText that contains personal identifiers such as names and email addresses
   • CDuplicate records that contain the same factual information
   • DText that is written in a language other than the primary training language
   Reveal Solution Discussion Select an answer first
4. 3application · medium

   A marketing agency is training a chatbot to generate product descriptions from customer reviews. The reviews contain a mix of positive and negative feedback, and some reviews include offensive language. The agency wants the chatbot to generate professional and polite descriptions, but also wants to capture the sentiment of the reviews. Which mitigation strategy is most suitable?

   • AExclude all reviews from customers who have used offensive language.
   • BRewrite the offensive reviews to remove the offensive language while preserving the sentiment.
   • CFilter out all reviews that contain offensive language.
   • DRedact the offensive words and replace them with a placeholder.
   Reveal Solution Discussion Select an answer first
5. 4foundation · easy

   A data pipeline contains a small number of records with toxic comments. The team wants to prevent these comments from influencing the model while preserving the rest of the dataset. Which mitigation strategy is most appropriate?

   • ARedacting the toxic words
   • BFiltering out the toxic records
   • CExcluding the entire dataset
   • DRewriting the toxic comments
   Reveal Solution Discussion Select an answer first
6. 5expert · hard

   A university is building a chatbot to answer questions about course offerings and campus events. The chatbot uses a knowledge base of emails and announcements that contain both public information and personal data of students (e.g., grades, disciplinary records). The personal data is scattered throughout the emails. The university wants to make the chatbot available to students, but must protect the privacy of individual students. Which mitigation strategy is most appropriate?

   • ARedact the personal data from the emails before indexing them for retrieval.
   • BExclude the entire emails that contain any personal data.
   • CUse a prompt that instructs the model to only answer questions about courses and events.
   • DAdd a response filter that blocks any output containing student names or grades.
   Reveal Solution Discussion Select an answer first