Hardest Google Cloud Generative AI Leader questions this week

Supervised learning is the correct approach because the company has labeled data (customers marked as churned or not churned) and wants to predict a specific outcome. The model learns the mapping from input features to known labels. Unsupervised learning would be inappropriate here because it is used when labels are not available, such as for clustering or anomaly detection without predefined categories.

Vertex AI Platform is Google Cloud's unified ML platform that supports the full ML lifecycle including data preparation, model training, deployment, and monitoring. BigQuery ML (A) is useful for running ML models directly on data warehouse tables but does not provide the full lifecycle management capabilities. Google AI Studio (C) is designed for prototyping with Gemini models rather than custom model training workflows.

Few-shot prompting provides the model with a small number of labeled examples directly in the prompt, which helps it understand the desired output format and classification scheme. This typically improves accuracy over zero-shot prompting (option A), which relies solely on the model's pre-trained knowledge without examples. Option C controls randomness but does not teach the model the classification categories, and option D only affects response length, not classification quality.