Zero Shot Prompting: Asking Without Examples

Chapter 2 introduces key terms and techniques frequently encountered in prompt engineering.

From zero-shot/few-shot prompting, familiar to anyone interested in AI, to RAG (Retrieval Augmented Generation), considered the next generation of search methodologies, you can explore a variety of terms and techniques related to prompt engineering.

Let's first explore the most fundamental technique in prompt engineering: Zero Shot prompting.

In AI learning, Shot refers to an example. Thus, zero-shot means handling a new task without examples, using only the pre-trained content by the AI without any special training.

Copy
Describe the difference between aerobic and anaerobic exercise in under 200 characters.

Copy
Tell me what the highest mountain in the world is and how tall it is.

In this way, making the AI interpret the prompt and derive results solely based on pre-trained data is termed Zero Shot Prompting.

Everyday prompts that do not provide specific examples are generally categorized as zero-shot prompts.

Note : Prompts that include some examples, like below, are known as Few-Shot prompts.

Copy
Here are some examples of news articles. Each article is categorized into a specific category.

Article: "Progress is being made in trade negotiations between countries. Economic experts view this positively."
Category: Economics

Article: "A famous actor has announced their role in a new movie. Fans are thrilled."
Category: Entertainment

Now categorize the following article.

Article: "Recent research suggests that a healthy diet can reduce the risk of heart disease."
Category: 

We'll discuss few-shot prompts in more detail in the next lesson.

• Quick Result Generation: It minimizes the time required to draft the prompt and allows the quick utilization of existing models for new fields or tasks without additional training.

• Flexibility: It is not confined to specific examples or data, allowing it to flexibly respond to various fields of work depending on the user's prompt.

• Accuracy and Expertise Issues: The accuracy of responses may be compromised in tasks from fields not yet learned.

• Time Waste Due to Ambiguity: If there are no clear guidelines, it could be difficult for the AI to provide correct answers. This may require more interaction between the user and AI, causing unnecessary time consumption.

Send prompt examples and compare the responses from the AI.