5minIntroduction to GPT's: Tips and Tricks
Recently, OpenAI released the functionality of its natural language generator to ChatGPT, which allows users to create custom conversational agents for their needs, called GPTs. With GPTs, users can make their own GPT with a specific context in mind, a knowledge base, actions, and different functionalities. Let’s quickly explore the process.
Name
The name for your GPTs is the first thing that the user will see when they interact with it. It should be catchy, memorable, and relevant to the topic or domain of your GPT. You can use emojis in the name to add some personality and flair to your GPT. For example, if you are creating a GPT for travel advice, you can name it something like “🌟🔮 Dreamy - Explain dreams” or “Dog Guru 🐕”.
Description - Your sales pitch
The description for your GPTs is a short summary of what your GPTs can do and why the user should chat with it. It should be clear, concise, and informative. You can use some keywords or phrases that relate to the topic or domain of your GPTs. For example, if you are creating a GPTs for fitness tips, you can write something like “Get fit and healthy with FitGPT 💪. This GPTs will give you personalized fitness advice, workout plans, and nutrition tips based on your goals and preferences.”. The description is limited to 300 characters, so think wisely.
Instructions - Prompt Engineering
You need to create a right prompt for your GPTs. The prompt should contain relevant information and definitions to the actions that GPT need to proceed The instructions for your GPT are the guidelines that you provide to the GPT on how to respond to the user’s queries or requests. The instructions should contain relevant information and definitions for the actions that the GPT needs to perform. You can use natural language, bullet points, or examples to explain the instructions to the GPT. For example, if you are creating a GPT for trivia games, you can write something like "This GPT is a trivia game master. It will ask the user trivia questions from various categories and keep track of the score. To create a trivia question, the GPT should follow these steps:
- Choose a category from the list: History, Geography, Science, Sports, Entertainment, etc.
- Choose a difficulty level from the list: Easy, Medium, Hard
- Generate a question and four possible answers, one of which is correct
- Mark the correct answer with an asterisk (*) at the end
- Format the question and the answers as follows:
Q: What is the capital of France? A) Berlin B) Rome C) Paris* D) London"
Image - Logo for your GPT
Here you need to create a unique logo for your GPT. The logo should represent the theme or domain of your GPT and catch the user’s attention. The image is in a circle and the dimensions of the viewable area are 80x80. You can use an online, or open source tool like to design your logo. Alternatively, you can use internal GPT’s Dalle-E tool to generate a logo based on a text prompt. For example, if you are creating a GPT for cooking recipes, you can write something like “vector image of a chef hat ”.
Conversation starters - GPT’s user manual
In this section you specify the list of the conversation starters for your GPT. These are a set of questions or instructions that the user will see when they open the chat with the GPT. The user can click on these questions to start a conversation with the GPT. The conversation starters should be relevant to the topic or domain of your GPT and invite the user to engage with it. We recommend having 4 questions. For example, if you are creating a GPT for movie recommendations, you can write something like:
- What genre of movies do you like?
- What is your favorite movie of all time?
- Do you want to watch a movie that is similar to or different from your favorite movie?
- Do you have any preferences for the actors, directors, or ratings of the movies?
⠀
Knowledge base - The brain of GPTs
A knowledge base is a collection of data that your GPTs can use to enhance its performance and accuracy. You can define and specify your knowledge base by uploading text files or other types of information, such as images, videos, or audio files. The GPTs will then use the knowledge base to learn from it, access it, and incorporate it into its outputs. It is very important to make a great preparation of the knowledge files, before submitting it. Taking into account that the context of GPTs have limited amount, having to much information might prevent you GPTs from working or might impact GPTs reasoning.
Capabilities - Smart processing of requests
GPTs can integrate some useful functions, that can perform various tasks based on natural language input. They can perform various tasks based on natural language input. They can get the content of the web page, generate images, analyze data, and execute code. Let’s explore them more in depth.
Web Browsing
This capability allows the GPT to interact with web content, such as browsing the internet, retrieving information, and interpreting web pages in real-time. This can help the GPT to answer queries, provide suggestions, or generate content based on the web sources. For example,
DALL·E Dalle-E is a GPTs capabilities based of Dalle-E3 that can create images from text descriptions by using web data as a reference. DALL·E is a GPT that can create images from text descriptions by using web data as a reference. However, DALL·E does not modify or edit existing images; it generates new images from scratch. Dall-E allows you to create the image. Beware, DALL·E generates new images from scratch; it doesn't modify or edit existing images. This is based on the limitation.
Code interpreter This capability enables the GPT to understand, write, and execute code in various programming languages. This can be useful for tasks related to software development, data analysis, and educational purposes. For instance, you can upload your code in the knowledge base and ask the GPT to run it, debug it, or optimize it. However, there are safeguards in place to prevent malicious use or unintended consequences of the code execution.
GPTs Actions - API communication
This functionality allows you to make third-party APIs available to your GPT by providing details about the endpoints, parameters, and a description of how the model should use them. You can also import actions for GPTs from an OpenAPI schema. This way, you can extend the functionality of your GPT by enabling it to call your API endpoints and retrieve information based on the user request and user input.
As you can see, building a complex GPT requires considering several aspects, such as prompts, creation, knowledge base, and actions. We have only touched the basics of these topics in this paragraph. If you need more help with creating custom GPTs, Dialog Duo can assist you with our expertise in these domains. Do not hesitate to contact us.
