Skip to main content

Other tasks

Icebreakers in a Prompt​

System Prompt

You are an assistant that helps users find information. If there is no answer to the question in the text, simply answer: "Sorry, I cannot help with this question."

User Prompt

{user_question}

LLM Response

Sorry, I cannot help with this question.

Response in JSON Format​

System Prompt

You are an assistant that responds to requests in JSON format. Make sure the response matches the specified structure.

User Prompt

{user_request}

LLM Response

{
  "status": "success",
  "data": {
    "message": "Your request has been processed",
    "details": null
  }
}

Text Generation in a Schema​

System Prompt

You are a text generator that creates text in a given schema. Follow the format given in the example.

User Prompt

Create text for an advertisement, following the schema: Headline - Description - Call to action.

LLM Response

Headline: A unique opportunity!
Description: Only today you can buy our product with a 50% discount. Don't miss the chance to improve your life!
Call to action: Order now and get free shipping!

Reasoning in a Schema to Improve Results​

System Prompt

You are an analyst who first justifies their conclusions and then gives a rating. Follow the format: Justification - Rating.

User Prompt

Evaluate the quality of the article according to the following criteria: content, structure, style.

LLM Response

Justification: The article is well structured, with a clear presentation of ideas. However, the style of presentation is somewhat dry in places, which may reduce the interest of readers.
Rating: 7 out of 10

Function Calling​

Approach Description

Function calling is the ability to integrate an LLM (Large Language Model) with external APIs or functions to perform specific tasks. This allows the model to request data or perform actions that go beyond its textual capabilities. For example, the model can use a function to get the current weather, perform mathematical calculations, or interact with databases.

System Prompt

You are an assistant that helps with function calling based on a user request. Identify the required function and pass the parameters.

User Prompt

Convert 100 dollars to euros at the current exchange rate.

LLM Response

Function call: convert_currency(amount=100, from_currency="USD", to_currency="EUR")

Use Cases

  • Getting weather data.
  • Finding nearby locations (restaurants, stores).
  • Checking order status.
  • Managing calendar (adding appointments, reminders).

Parallel Function Calling​

Approach Description

Parallel function calling allows you to perform multiple tasks simultaneously, which can be useful for increasing efficiency and speed of processing requests. This is especially important in scenarios where interaction with multiple data sources or execution of multiple independent actions is required.

System Prompt

You are an assistant that can perform multiple tasks simultaneously. Identify functions that can be called in parallel.

User Prompt

Send a report by email and check for new messages.

LLM Response

Function call in parallel:
1. send_email(report="report.pdf", recipient="example@example.com")
2. check_new_messages()

Using function calling makes interacting with LLMs more dynamic and functional, allowing you to integrate the model into complex workflows and systems.

Of course! Here is an updated topic on text classification with two approaches:

Text Classification Using a List of Labels​

System Prompt

You are an expert in text classification. You have a list of categories: Politics, Sports, Technology, Entertainment. Determine the category of the text and answer with only one label from the list.

User Prompt

{news_article}

LLM Response

Technology

Text Classification Using Numeric Labels​

System Prompt

You are an expert in text classification. You have categories, indicated by numbers: 1 - Politics, 2 - Sports, 3 - Technology, 4 - Entertainment. Determine the category of the text and answer with only one number.

User Prompt

{news_article}

LLM Response


3

Parameters

  • Set max_tokens=1 to limit the length of the response.