1.2 Describe How Various Data Types Are Used in Gen AI and the Business Implications. GENERATIVE-AI-LEADER Practice Questions (Page 2)

Questions 6–10

1. 6application · medium

   A government agency is developing a gen AI model to summarize public comments on proposed regulations. The comments are collected from various online platforms, and some are duplicates, some are off-topic, and others contain contradictory statements. The agency has a limited budget and needs to ensure the model produces reliable summaries. What should they do first?

   • ADeduplicate the comments and filter out off-topic submissions to improve the relevance and consistency of the training data.
   • BUse all comments as-is to avoid bias in the model, since the public's voice should be fully represented.
   • CManually review and edit each comment to ensure completeness and consistency.
   • DConvert all comments to a structured format like CSV to make them easier to process.
   Reveal Solution Discussion Select an answer first
2. 7foundation · easy

   A company has a large dataset of product images but no annotations. They want to use this data to pre-train a generative AI model that can create new product images. Which statement best describes the role of this unlabeled data?

   • AIt cannot be used at all for AI training because labels are always required.
   • BIt must be manually labeled before any model can use it.
   • CIt is only useful for testing, not for training.
   • DIt can be used for unsupervised pre-training to learn patterns and structures in the images.
   Reveal Solution Discussion Select an answer first
4. 8expert · hard

   A retail chain wants to use gen AI to generate product recommendations based on customer purchase history and social media posts. The purchase history is stored in a structured database, while social media posts are unstructured text. The company has a limited budget and needs to decide how to prioritize data collection and processing. Which approach is most cost-effective?

   • AOutsource the processing of social media data to a third-party service to reduce the burden on the internal team.
   • BInvest heavily in collecting and processing social media data, as it provides richer insights into customer preferences.
   • CUse both data types equally from the start, as a diverse dataset will yield the best model performance.
   • DUse the structured purchase history first, as it is readily available and requires minimal processing, then gradually incorporate social media data as budget allows.
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A logistics company wants to use gen AI to analyze customer feedback from surveys (numeric ratings), emails (free text), and phone call transcripts (audio converted to text). They have a limited budget and need to decide which data types to prioritize for initial model development. Which data type should they prioritize to achieve the quickest results with the least preprocessing effort?

   • ANumeric survey ratings, because they are structured and require minimal preprocessing.
   • BPhone call transcripts, because they provide the most detailed feedback from customers.
   • CFree-text emails, because they contain rich customer sentiment that numeric ratings cannot capture.
   • DCombine all data types equally, because a diverse dataset will yield the best model performance.
   Reveal Solution Discussion Select an answer first
6. 10expert · hard

   A tech company is developing a gen AI model to generate code snippets from natural language descriptions. They have a large dataset of code repositories, but only a small portion is labeled with the corresponding natural language descriptions. The team is considering using a pre-trained model and fine-tuning it. They have a limited budget and need to decide how to allocate resources between labeling and compute. Which strategy is most effective?

   • AAvoid using a pre-trained model to reduce costs, and instead train a simple model on the labeled data.
   • BUse only the unlabeled data to train a model from scratch, as it is more cost-effective than labeling.
   • CInvest in labeling a larger portion of the data to improve the model's accuracy, even if it means using a smaller pre-trained model.
   • DUse a large pre-trained model and fine-tune it with the limited labeled data, leveraging the model's existing knowledge of code and language.
   Reveal Solution Discussion Select an answer first