IS 4490 - Prompt Engineering and Learning Techniques

What are the four key components of prompts for AI models?

Instruction/Intent, Context, Reasoning Assistance/Guidance, Data

What does the "Context" component of a prompt provide?

The situational backdrop and environmental information

Which type of reasoning assistance provides decision policies such as decision trees or if-else rules?

Rule-based reasoning

What is the core principle of metadata prompting?

To separate the task description from entity explanations

In metadata prompting, how are entities distinguished within the task description?

They are enclosed in backticks (`)

What happens to the model's internal weights during few-shot prompting?

They do not update at all

What distinguishes many-shot prompting from few-shot prompting?

Many-shot uses several hundred examples instead of just a few

What is a significant downside of many-shot prompting?

It increases computational cost and slower inference times

What does "In-Context Learning" refer to?

A model's ability to perform tasks by interpreting examples in the prompt without updating parameters

Which of the following is NOT mentioned as a benefit of zero-shot prompting?

Improved computational speed

What should be ensured when selecting exemplars for few-shot prompting?

Exemplars should follow a consistent format and structure

What is Retrieval-Augmented Generation (RAG) used for?

To dynamically retrieve relevant data from knowledge bases

Which output format is often most common for AI-generated data that will be used by downstream tasks?

JSON/XML

According to the material, what does Data in a prompt represent?

The specific input material the model needs to process, analyze, or transform.

How does Metadata Prompting primarily differ from traditional prompting methods?

It separates the task description from entity explanations, defining entities in a structured format.

Which of the following describes In-Context Learning?

The model interprets examples in the input prompt to perform a task without updating its internal parameters.

What is the primary benefit of Zero-Shot Prompting?

It allows models to handle a wide range of tasks without fine-tuning or retraining.

What is a disadvantage of Many-shot Prompting?

It can increase computational cost and slow down inference times.

The "Instruction/Intent" component should specify both what needs to be done and exactly how to achieve it.

False

Few-shot examples can use angle brackets (<>) to create placeholders for information that should be inserted.

True

In metadata prompting, entity explanations are described separately in JSON format after the task description.

True

The model permanently retains information learned from few-shot examples after the interaction is over.

False

Steps-based reasoning provides sequential steps the AI should take to reach the final output.

True

Metadata prompting makes prompts more cluttered and harder to understand.

False

In-context learning includes both few-shot and many-shot prompting as forms of this approach

True

Constraints in reasoning assistance specify what the model should do rather than limitations on output generation.

False

The number of exemplars used in few-shot prompting should always be maximized regardless of task complexity.

False

Markdown format is particularly useful for generating tables and content requiring semantic structure.

True