Gen AI Quiz 1

A retail company wants to predict customer churn (is customer likely to leave us) based on past purchase behaviour, customer service interactions, and demographic data. Which type of machine learning problem is this?

• Unsupervised learning - clustering

• Supervised learning - classification

• Supervised learning - regression

• Reinforcement learning

• Not familiar with this

A business unit wants to use machine learning to automatically group similar customer support tickets so they can be routed to the correct specialized team faster. No pre-labeled examples of these groupings exist yet. Which type of Machine Learning is best suited for this task?

• Supervised Learning (Classification)

• Supervised Learning (Regression)

• Unsupervised Learning (Clustering)

• Deep Learning (Reinforcement)

Your SAP system needs a new feature: a Copilot that drafts detailed, context-aware procurement contract clauses based on a brief, natural language request (e.g., "Draft a new penalty clause for late delivery in Brazil"). Which technology is appropriate to handle this task?

• Classification Model (Machine Learning)

• Large Language Model (LLM) / Generative AI

• Traditional NLP (NER/Topic Modeling)

• Regression Model

Transformer models have unique features. Select all that apply (2 correct answers).

• Attention layers

• Recurrent connections

• Convolutional layers

• Feedforward neural networks

You are trying to get a Generative AI model embedded in SAP to summarize a complex, 10-page legal document into an action-oriented memo for a non-legal manager. Which of the following is an example of an effective prompt engineering technique for this task?

• "Summarize the document."

• "Tell me everything about the document."

• "Act as a legal analyst for a risk-management VP. Summarize the attached document into a bulleted memo of no more than 5 points, highlighting only the three most significant risks and the required next action steps for the business unit."

• "Why is this document important?"

Which of these activities is a core Generative AI capability that fundamentally creates new, original content to increase business efficiency, as opposed to simply analyzing existing data (Predictive/Descriptive ML)?

• Predicting whether a piece of equipment will fail in the next 30 days

• Generating personalized marketing emails and product descriptions based on customer segment data

• Classifying vendor invoices as approved or rejected

• Clustering customer profiles for segmentation analysis

Which of the following are key differences between traditional machine learning and Generative AI? (three correct answers)

• Generative AI can create new data; traditional ML focuses on predictions

• Traditional ML is trained with labelled data whereas Generative AI works with unlabelled data

• Generative AI models need a lot more data for training compared to machine learning models

• Traditional ML methods predict based on patterns in the data unlike Generative AI

• Traditional Machine Learning models are more explainable than Generative AI models

What’s the difference between tokens and parameters in the context of Large Language Models (LLMs)? (2 correct answers)

• Tokens are the input data chunks the model processes, while parameters are the model's learned weights

• Tokens determine the model’s accuracy, while parameters determine the model’s speed

• Tokens are fixed for all models, while parameters vary based on the architecture

• Tokens are part of the training process, while parameters are used only during inference

Why is using a well-crafted system prompt often recommended when building solutions that use LLMs? Select three answers.

• The system prompt sets high-level behavior and role for the LLM.

• The system prompt can serve as a guardrail to prevent unintended use.

• The system prompt overrides user instructions whenever a conflict arises.

• The system prompt helps provide consistent context across multiple user requests.

A Business Unit wants to allow employees to search internal policy documents using natural language queries (for example, "What is the policy for expense reporting in EMEA?"). What approach will help find the most relevant answer from the policy documents?

• Convert the documents to binary codes; convert the search query to a binary code and search in the converted documents.

• Match the query's text against the text in the policy documents.

• Upload the document to the LLM along with the question so the LLM can reason and respond.

• Convert the documents to embeddings and store them in a vector database; convert the query to embeddings and perform a nearest‑neighbor search on the vector database.