How to Use OpenAI’s GPTs to Create Your Own Custom Chatbots 

On November 6, in San Francisco, OpenAI hosted its first Dev Day event, introducing GPTs.

GPTs represent a major leap in customization, as they allow users to tap into the potential of the GPT Builder by simply providing natural language instructions. This empowers users to craft their very own chatbot tailored to their specific needs and tasks.

Furthermore, GPTs have the ability to seamlessly integrate with third-party service APIs, greatly expanding their functionalities.

OpenAI has also excitingly revealed their plans to launch the GPT store later this month, providing a platform for users to market and sell their custom chatbots.

In this article, we provide you with a comprehensive guide on how to create GPTs using the GPT Builder and introduce a simple chatbot we’ve developed to assist users in identifying phishing emails.

What is a GPT, and How Can You Create One?

GPTs essentially represent customized iterations of ChatGPT, each bearing its distinctive name and logo, along with a predefined set of prompts that users can readily access to interact with the virtual assistant.

As an illustrative example, OpenAI introduced the Canva GPT tool, which offers users a set of four predefined prompts. These prompts cater to various tasks, such as crafting an inspirational quote for social media, generating an Instagram post, designing a poster, and outlining details for a Facebook post featuring a hiking trail

Users now have the opportunity to craft their very own GPT tools using the GPT Builder, a virtual assistant developed by OpenAI for this precise purpose.

Within the GPT Builder interface, users can input written instructions that outline the specific functionalities they desire for their custom GPT. Additionally, they have the flexibility to assign a distinctive name and logo to their creation, further personalizing the tool to their preferences and needs.

Creating your own GPT is a streamlined process, and you can get started quickly by following our fast-track list. (More detailed step-by-step instructions are provided in the following section.)

How to Create a GPT: The Fast-Track TLDR Explanation

1. Purchase a ChatGPT Plus or ChatGPT Enterprise Subscription through this link here.
2. Open ChatGPT.
3. Navigate to the left sidebar and click on the Explore option.
4. Choose the Create a GPT option.
5. Begin by entering your initial prompt or instructions, detailing what tasks you want your GPT to perform.
6. Engage with the agent by responding to any follow-up questions it poses.
7. To ensure your progress is saved, click on the Save icon located in the top right corner.

Getting Started With the GPT Builder

Before diving into the GPT creation process, it’s essential to have either a ChatGPT Plus or Enterprise subscription. These premium subscriptions begin at a cost of $20 per month and can be acquired by visiting the provided link here.

To complete the purchase, follow these steps:

1. Scroll down to the Compare ChatGPT plans table.
2. Locate the Plus section and click on the Upgrade now button.
3. Proceed by either signing up with a new account or entering your payment details to finalize the subscription process.

Once you’ve successfully obtained a paid subscription, access ChatGPT on the OpenAI website using the provided link.

Then, follow these steps:

1. Visit the OpenAI website and use the link provided to open ChatGPT.
2. Once the interface is open, locate the black sidebar at the top-left corner of your screen.
3. Click on the Explore option found within the sidebar (as illustrated in the top-left section of the screenshot above).

Upon clicking Explore, you’ll notice that your screen transitions to a list view with two sections: My GPTs and Made by OpenAI.

• My GPTs presents a list of GPTs you’ve previously created.
• Made by OpenAI contains a list of GPTs developed by OpenAI.

Under the My GPTs list, click on the Create a GPT option to launch the GPT Builder interface, where you can begin the process of crafting your own GPT tool.

How to Use the GPT Builder

Once the GPT Builder is open, you’ll be directed to the tool’s Create pane, where you can enter your instructions to develop your GPT.

At the bottom of the page, you’ll encounter a dialog box prompt labeled Message GPT Builder. Here, you can enter instructions that convey to GPT Builder the specific tasks and functionalities you want your GPT to possess. (You will find a more detailed breakdown of what kind of instructions you can enter to create a GPT in the next section).

Additionally, you can inquire about the chatbot’s capabilities by asking questions.

Adjacent to this dialog box, you’ll find a Paperclip icon. Clicking on this icon enables you to upload various attachments, such as Word documents or Excel files. This feature provides you with the capability to feed data or images directly into your GPT chatbot.

It’s important to mention that you also have the option to construct a GPT through the Configure tab, which can be accessed by clicking on the Configure option located in the top left corner of the screen.

When you access the tab, it opens a new view where you can input a Name, Description, and Instructions for your chatbot. Additionally, you have the option to include Conversation Starters, which are essentially prewritten prompts that users can click on when using your GPT. This feature streamlines user interactions, eliminating the need for users to start from scratch.

Once you’ve completed the development of your bot, simply click on the Save icon located in the top right corner of the screen to preserve your progress.

In addition, there is an arrow included with the option that allows you to save your new GPT in one of two ways:

• As a private model, restricted to your ChatGPT Plus or Enterprise account,
• or as a private link that you can share with others.

These settings can be changed at any time.

Techopedia’s Phishing Awareness GPT: PhishGuard Assistant

In this section, we’ll explore a basic GPT that we’ve developed called PhishGuard Assistant. The goal for this exercise was to create a basic GPT capable of scanning and identifying phishing emails.

Disclaimer: It’s crucial to emphasize that this exercise is purely theoretical. We strongly recommend that enterprises conduct their independent research to determine the reliability and effectiveness of large language models (LLMs) as a solution for detecting phishing emails.

Now that we’ve clarified the context, let’s discuss how we did it. For the scope of this exercise, our goal was to construct a GPT Bot that an employee could employ to scrutinize emails, identifying potential phishing or scam messages.

To initiate the process, we began by entering a prompt that outlined the GPT’s purpose. Additionally, we included four conversation starters that users could readily click on for interaction.

Here’s a screenshot of this initial prompt:

In the screenshot provided, you can see that GPT Builder responded after approximately a minute of processing. It inquired about the name we wanted to assign to our GPT and offered a mock layout of it in the Preview section on the right-hand side.

Subsequently, we requested GPT Builder to name our new GPT “PhishGuard Assistant.” GPT Builder complied by incorporating the chosen name into the Preview pane and proposing a profile picture.

Following a minor adjustment request, we accepted the revised version.

A screenshot of this exchange can be seen below:

Below is the revised logo, and response the GPT Builder generated:

GPT Builder not only updated the logo but also, in its initial response, sought guidance on the type of responses it should generate for users.

In response to this, we provided instructions for the GPT to blend technical details and general characteristics in its responses. We also specified that it should offer explanations that are accessible and comprehensible to a wide range of users.

GPT Builder proceeded to seek our input on how it should handle incomplete user input, to which we directed it to request further clarification from the user.

It also inquired about the preferred tone of voice for responses. We opted for a tone that strikes a balance between professionalism and friendliness.

At the conclusion of this interaction, GPT Builder confirmed that PhishGuard Assistant was ready for us to explore in the Preview pane.

In the following section, we’re going to look at how it works.

Testing the PhishGuard Assistant GPT

To initiate our testing of PhishGuard Assistant GPT, we selected the “Is this a phishing email” conversation starter.

It generated the following response:

In response, we provided an example of a phishing email by copying and pasting content from a source on the UC Berkeley website, as indicated in the link.

PhishGuard Assistant, in turn, delivered a comprehensive breakdown of red flags within the email. This detailed analysis offers valuable insights for users seeking to validate the legitimacy of an email, aiding them in identifying potential phishing attempts.

A screenshot of this is presented below:

This input appeared to be relatively detailed and demonstrated that the prompt was functioning more or less as intended.

Clicking on the other conversation starters also provided valuable guidance on preventing phishing attacks and presented a realistic example of a phishing email, including details highlighting its manipulative aspects.

At a broader level, this exercise showcases how GPT Builder can effectively be employed to create a chatbot designed to address recurring queries and provide informative responses.

The Bottom Line

This article has merely touched upon the potential of what you can achieve with the GPT Builder. If you’re aiming to construct a more intricate bot, it’s essential to invest time and exercise patience, considering that GPT Builder may take some time to respond to your input.

The good news is that if you ever find yourself uncertain about how to navigate and utilize GPT Builder, you can seek assistance from it at any point, ensuring a smoother and more productive experience.