🆕 OpenAI: GPT-3, ChatGPT and GPT-4

OpenAI offers GPT-3, ChatGPT and GPT-4. The GPT-series uses state-of-the-art NLP technology to generate responses in your chatbot, by providing answers that fit the context of the conversation.

The GPT-3 integration is deprecated in favour of this new OpenAI integration which contains GPT-3, ChatGPT and GPT-4, and potential future APIs offered by OpenAI.

This documentation page covers:

1. What is GPT

2. [Video] GPT in customer-facing conversations

3. How to set up your GPT integration

4. How to use GPT inside a bot

   1. [Use-case] GPT for fallback messages

   2. [Use-case] GPT for contextualized responses

5. Limitations of GPT

What is GPT

GPT is a series of advanced Natural Language Processing systems developed by OpenAI that allow you to execute several text-based tasks.

The OpenAI integration allows you to generate responses in your chatbot. As a bot builder or conversation designer, all you need to do is select the GPT-3, ChatGPT or GPT-4 model and tell it the right query that defines the kind of answer that you would like, and it will make it up for you!

To be able to use GPT models on Chatlayer, you will need to set up the OpenAI integration, then use this integration inside bot dialogs.

Using the OpenAI integration means that your data is sent to OpenAI services which are based in the United States. Make sure you understand the OpenAI terms of service and implications. Use OpenAI in production bots with caution, Chatlayer is not held responsible for generated content.

[Video] GPT in customer-facing conversations

Here is a video we made about how and why you could bring value to your bot if you use GPT.

5MB

How to use GPT integration inside your Chatlayer chatbot.mp4

How to set up your OpenAI integration

Before being able to use GPT inside dialogs, you need to connect an OpenAI account to the OpenAI integration on Chatlayer. For this, you might require an OpenAI API key.

Chatlayer provides customers with its own OpenAI API key to use GPT with a limit of 100 messages per bot. This means that from the 101st message, GPT will stop generating and send errors. If you wish to be able to use an unlimited amount of GPT-generated messages, you will need to create your own API key on the OpenAI's website.

To connect your chatbot with GPT:

• On your Chatlayer home page, go to Integrations

The Integration tab is on the left-hand side of your Chatlayer home page.

• Search for "OpenAI" in the search bar

Search for OpenAI and you will find the OpenAI integration ready to be configured

Click on the integration to open it

View of the OpenAI integration when open.

• Click on + Add Account to start connecting your OpenAI account with the OpenAI integration

• This will open a pop-up in which you will need to enter a Chatlayer Access token:

Pop-up to provide the necessary credentials for connecting OpenAI to Chatlayer

• Paste your access token in the Access token field

If you do not have a Chatlayer access token, you can generate one here.

• Under Display name, write an identifier that can be later used to select this specific access token when using the OpenAI integration inside a bot. In this example, we have used the identifier Chatlayer Token OpenAI Example.

• Optionally, you can provide an OpenAI API key.

To try the OpenAI integration, Chatlayer offers 100 messages per bot. If this suits you, you can leave the OpenAI key field empty. Yet if you would like to go unlimited in the your number of GPT-generated messages, you will need to get your own API key on the OpenAI's website.

• Click on Submit when you're done

• The OpenAI account should now be visible under the Connected Accounts tab. Note that the word "Auth" was automatically added in front of the "Display Name".

You can see your OpenAI account under the Connected Accounts tab.

You've set up your GPT account and would like to jump straight to bot building? We've got you covered with our GPT-3 template bot. Give it a go!

Your OpenAI integration is set up! Now it's time to use it inside the conversation.

How to use OpenAI GPT models inside a bot

You will need to do two things so that your customers can interact with GPT-3, ChatGPT or GPT-4:

1. Generate a GPT response through an Action dialog

2. Display a GPT response within another dialog

Generate a GPT response

To generate a GPT-3, ChatGPT or GPT-4 response:

• Create an Action dialog and open it

• Search for the OpenAI integration in the search bar and select it

Search for the OpenAI integration in the Action dialog toolbar and select it.

• Under OpenAI account, select your OpenAI account that you connected to Chatlayer earlier. In our case, it is called "Auth Chatlayer Token OpenAI Example".

Enter your OpenAI account to connect it to the Action dialog.

• Select the action to execute, called "GPT response". This means the action will send the conversation to OpenAI and retrieve a response.

Select the action for the GPT-3 integration.

• Now, give to OpenAI the required information to be able to generate an answer. These are called the Action fields:

  • Fill the Bot description, which is a small depiction of what your bot is about. For instance, a restaurant bot.

  • Fill the Bot prompt, which is a description of the response that you would like GPT to generate. Example: generate a personalized message to apologize for an inconvenience.

  • Select the Model Type, which is either GPT-3, ChatGPT or GPT-4. Note that GPT-4 is only available in Chatlayer US due to European GDPR regulations.

Be mindful: the longer you make your bot description or bot prompt, the longer it will take to get a response from OpenAI.

Fill the Action fields so that OpenAI gets all the necessary information to generate a response.

• Click on Create to save your GPT integrated Action dialog

You have made an Action dialog that uses GPT to generate a response based on your chatbot description! Now, let's see how we can display that generated response within the conversation.

Display a GPT response

The generated answer from GPT is available on the session data after executing the above Action dialog. This response is saved under the {apps.openai.gpt.text} variable.

To display the GPT response in the conversation:

• Create a Bot message dialog

• Add a text field and fill it with the{apps.openai.gpt.text} variable

Add the variable that holds the generated response from OpenAI to a text field to display it in your conversation.

You know now how to set up an Action dialogs that generates a GPT response, and how to display this response within a dialog.

In the next sections, we will take a use-case to exemplify how you can use GPT for your bot.

[Use-case] GPT for fallback messages

One way to use the OpenAI integration is as a fallback message for when your bot didn't understand the user question.

If you do not have a bot yet, you can use one of our pre-built template bots.

Once you've created a bot, go into your General flow, and open the Not Understood dialog. From there, we will modify the Not Understood dialog so that it uses the OpenAI integration:

• Delete the text message inside the Not Understood dialog by clicking on the bin icon

• In the Go To field, create a new Action dialog that you will call "Generate GPT response", then save your changes

• Open the Generate GPT response dialog that you have just created and configure this Action dialog so that it integrates GPT

• In the Go To field, create a Bot message dialog and call it "Display GPT response" , then save your changes

• Open the Display GPT response dialog and add the {apps.openai.gpt.text} variable inside the text field

How your fallback flow should look like if you use GPT as a Not Understood message

If you talk to your bot now and ask something the bot doesn't know, the "Not Understood" dialogue state should be triggered. Next, OpenAI will be called and the response will be displayed to you.

Example of a fallback flow where GPT-3 was asked to generate an apology for an inconvenience in a restaurant bot.

[Use-case] GPT for contextualized responses

GPT is able to read the whole conversation between your customer and the bot, which means that it can generate text based on the context of the conversation.

As for an example of contextualized GPT responses:

• Create an intent that will trigger a GPT response. In this example, we created an intent_not_delivered intent.

• Create an action dialog that generates a GPT answer and connect it to a dialog that displays it, as explained in the tutorial above. In our example, we have queried the following text to GPT-3:

Example of action fields for a personalized apology response made by GPT-3

• Do the same for any other dialog that you would like to have personalized. In our example, we have created a chitchat.tell_a_joke intent that triggers a GPT-generated response of a joke that takes into account the context above:

• 

  Example of action fields for a personalized joke made by GPT-3 and based on the context of the conversation
• Your can now have more contextualized interactions with your customers!

Conversation sample of a flow that uses GPT-3 to generate a personalized apology then a joke

Limitations of GPT models

GPT-3, ChatGPT and GPT-4 are generative AI models built by OpenAI, which is located in the US. Because it is generative, we do not control what GPT creates. Depending on the input it receives, the answer can either be very good, contain wrong info or even be offensive. Therefore, we do not recommend GPT be used in production (i.e. a customer facing bot) without thorough testing and understanding the risks.

We do not recommend the OpenAI integration to be used in production (i.e. a customer facing bot). Still, if you want to use it in production anyway you should use your own OpenAI API key. Please also check the OpenAI terms of service to understand the implications of using OpenAI services.

Feedback?

We are eager to receive your feedback if you found a valuable and safe production use case. Please contact us here. 🙏