Multi-Shot (Multiple Examples)

TLDR: Multi-shot prompting, also known as few-shot prompting, is a technique used in generating prompts for ChatGPT. It involves providing multiple example inputs or prompts to guide the model's responses. By exposing the model to a few different prompts, it can learn and generalize patterns, enabling more context-aware and diverse responses.

Also known as: Single-shot, Few shot, multiple examples

What is a Shot?

A shot is used in prompt engineering and it means 'examples' or 'prompt inputs'.

What is Multishot?

Multi-shot means multiple examples. It involves providing multiple examples in a prompt to train an LLM to respond in a structured manner. The structure it responds with is defined in the examples/shots provided.

When should you use it?

Suppose you want ChatGPT to summarize multiple analyst reports authored by different banks. You want the summary output to look in a particular way, e.g., [Bank 1] - [Rating - Sell/Buy], [Bank 2] - [Rating - Sell/Buy], etc.

In your prompt, you will then provide examples of the desired structure to guide ChatGPT into generating outputs based on those examples.

Common scenarios where it is utilized

• Technical translation

• Generating summaries/content based on a set structure

• Customer service interactions

Examples

Translation

If you are developing a language translation system that aims to accurately translate technical documents across various disciplines. By providing the AI with multiple examples of well-translated technical documents, you enable it to grasp the nuances of different terminologies and domain-specific language.

Prompt: You are going to translate technical text from English into Chinese using technical Chinese in the translation.

I will provide three examples of a technical paragraph in English and its corresponding translation in Chinese. Follow a similar structure when translating a new paragraph.

Paragraph 1: English Text

Translation: Chinese Translation

Paragraph 2: English Text

Translation: Chinese Translation

Paragraph 3: English Text

Translation: Chinese Translation

Now Translate the following document: Document to translate

Specific Structure

Let's explore a different scenario: imagine you are developing a news summarization system that needs to compile concise summaries of news articles. To train the AI in generating structured summaries, you can provide a few examples of well-structured summaries in the desired format. This allows the AI to learn the underlying structure and apply it to summarize new articles effectively.

Prompt: You are going to summarize news articles into concise summaries I will provide three examples of news articles and their corresponding summary. Follow a similar structure when summarizing different news articles.

Article 1: Text

Summary: Article Summary

Article 2: Text

Summary: Article Summary

Article 3: Text

Summary: Article Summary

Summarize the following news article: Article to summarize