¶ 1. Knowledge base
Go to Knowledge page and create an instance of knowledge
GPT System Instructions are a set of predefined rules, guidelines, or constraints designed to shape the behavior, tone, and functionality of a GPT-based model (like ChatGPT). These instructions help the model understand how it should interact with users, manage different conversational scenarios, and provide responses that align with the intended use case.
Key Aspects of System Instructions:
Behavioral Guidelines: Define how the model should behave—e.g., formal or informal tone, level of detail in responses, or how much creativity to apply.
Role Definition: Specify the model's "persona" or "role." For instance, the model might act as a technical assistant, a creative writer, a tutor, or a customer support agent.
Ethical Boundaries: Impose restrictions to ensure the model operates responsibly, avoids harmful or inappropriate content, and adheres to ethical norms.
Functionality Scope: Limit or expand the topics the model can discuss, tools it can access, or actions it can perform.
User Context Management: Provide guidance on maintaining continuity in conversations, remembering relevant information, or adapting responses to a user’s preferences.
Fallback Behavior: Define what the model should do if it encounters ambiguous, incomplete, or unsupported queries (e.g., asking for clarification or gracefully declining).
Example of a System Instruction:
"You are an empathetic and knowledgeable virtual assistant. Respond concisely and accurately to user queries while maintaining a friendly and supportive tone. Please do not answer questions that are not related to our services."
These instructions serve as the foundation for ensuring that the GPT model behaves consistently, aligns with the goals of its deployment, and meets user expectations.
The GPT chatbot generates its answers by utilizing the knowledge base, which is equally enriched by information from the "Base Information," "FAQ," and "Files" sections. Each section is treated with equal importance and weight in contributing to the chatbot's response formulation, ensuring comprehensive and balanced answers.
Do not give contradictory instructions, as artificial intelligence will behave unpredictably.
For example, if you tell the Assistant:
"You do not engage in any other type of conversation other than chatbots and live chat systems for customer support."
And then you provide the following information in the Knowledge Base: "The company founder is a fan of surfing, rock climbing, and working on chatbot architecture."
The assistant is likely not to provide information about surfing and rock climbing, as it is prohibited from communicating about anything other than the chatbot.
! The knowledge base serves as the foundation for the Assistant's responses. The Assistant does not reply with memorized phrases. Instead, it generates responses independently, adapting them according to information in the knowledge base and influenced by its configuration settings.
For example, the Assistant's tone and style can vary depending on its setup, such as responding in a formal, professional manner or adopting a friendly, cheerful approach.
¶ 2. Assistants
Go to Assistants page and create an instance of assistant
Please pay attention to the status of your Assistant. The status should be successful. If the status is "In queue", then reload the page to update the status information. If the status is failed, then open the assistant to view the error.
¶
3. Open AI key
To create an Open AI key, an organization must first be created.
Creating an organization in OpenAI involves setting up a workspace for team collaboration, managing billing, and organizing projects. Here’s how you can do it:
Steps to Create an Organization in OpenAI:
Sign Up or Log In: Visit OpenAI’s website and log in to your account. If you don’t have an account, create one.
Access Organization Settings: Once logged in, navigate to your account settings. Look for a section labeled "Organization" or "Manage Organization."
Create a New Organization: Click on the option to create a new organization.
Invite Team Members: Add team members by entering their email addresses. Assign roles such as admin, developer, or viewer, based on their responsibilities.
Set Up Billing: If your organization plans to use paid features or API credits, configure the billing settings. OpenAI accepts various payment methods for subscriptions or usage-based charges.
https://platform.openai.com/settings/organization/billing/payment-methods
Configure API Keys: Generate and manage API keys for your organization. These keys allow your team to integrate OpenAI's API into projects securely. Create an Open AI key on this page and add to Assistant.
! Please note that the API key must be created in the organization, not the user's API key.
¶
4. What are Temperature, Model, and Top P?
Temperature:
- Purpose: Controls the randomness of the Assistant's responses.
- Low Values (e.g., 0.2): The output is more focused and deterministic, ideal for tasks requiring precision or consistency.
- High Values (e.g., 0.8): The output becomes more creative and diverse, suitable for brainstorming or open-ended discussions.
Model:
- Purpose: Refers to the version or architecture of the AI model powering the Assistant.
- Each model may vary in capability, response quality, and resource efficiency. Newer models typically offer improved understanding and response generation.
Top P (Nucleus Sampling):
- Purpose: Adjusts the range of response possibilities by considering a subset of likely outcomes.
- Low Values (e.g., 0.1): The output is focused on the most probable responses.
- High Values (e.g., 0.9): The output includes a broader range of possibilities, allowing for more varied and creative responses.
By tuning these settings, users can customize the Assistant's behavior to better match their specific needs.
¶ 5. Knowledge Base editing
To add new information to the Knowledge Base and ensure the Assistant uses the updated data, follow these steps:
- Edit the Knowledge Base with the new or revised information on your promptchat White label website.
- Navigate to the Assistant Instance on the Assistants Page on your promptchat White label website.
- Click the "Save" button to initiate synchronization.
Once saved, the Assistant will automatically sync with the updated Knowledge Base, allowing it to use the latest information in its responses.
¶
6. AI Assistant in Chatbot
Create a chatbot on the Chatbots page and let's take a closer look at the gpt chatbot structure:
(You can download an example of a template for such a chatbot at the following link /gpt_functionality/gpt-ng-chatbot.zip )
A question block should be in front of the AI Assistant, which will contain the first question for the Assistant.
This is an example of what your chatbot could look like.
Ensure that one or more blocks precede the AI Assistant block. These blocks will serve as the basis for the chatbot to generate its response. You can adjust the number of these blocks considered by the chatbot in the "Count of history messages" field.
Client data is the information you want to pass to the chatbot as additional information about the client.
You must select an assistant instance that was created previously.
The AI Assistant block has an auto-tagging function. The AI chatbot adds tags to a conversation when it understands that the conversation was related to one or more of the tags.
Ways to exit the AI assistant block.
Use the "Next steps" tab to define and configure the actions that will trigger the chatbot's exit. Clearly describe these steps to ensure the chatbot understands the conditions under which they should be applied.
The “Error Step” exists by default and does not need a description. Exiting using the "Error step" will be done if we receive an error from the Open AI Platform.
If the client has an "Error" exiting the “Assistant block”, the following logic can be configured for the chatbot flow:
- create a “Message block” and offer the user to contact the “Assistant block” again by rephrasing their question. (Example 1)
- create a "Buttons block” and offer the user to contact the“Assistant block” again or choose other options to continue the flow. (Example 2)
- create your own version of continuing the flow
Example 1
Example 2
Before using a chatbot with an AI Assistant, we recommend testing it with different questions to see what answers you get. And if necessary, edit the instructions.