The Set System Instructions node is a fundamental component for configuring how your AI models behave. This guide explains how to use system instructions effectively in your Neurons.

​
Functionality

The Set System Instructions node allows you to define or modify the System Instructions for a Neuron.
The last Set System Instructions node set in your workflow is the instructions that LLM will use.

​
Node Properties

  • System Instructions:
    • The system instructions to be used for the LLM.
  • Prompt Position: Select how you want the instructions to behave:
    • Insert Text Before Existing System Instructions: Add text at the start of any existing system instructions. This can allow you to inject a short introductory statement, or additional instructions.
    • Insert Text After Existing System Instructions: Add text to the end of the existing system instructions. This position could be useful for providing clarifying information or examples.
    • Replace Entire System Instructions: Overwrite the entire system instructions with a new text, effectively resetting the LLM’s context.

​
Usage Examples

​
Scenario: Defining an AI assistant

In a customer support chatbot neuron, you might use a Set System Instructions with a prompt that reads: 
You are a helpful and concise AI customer support assistant. Your goal is to answer customer questions politely and effectively.

​
Tips and Best Practices

  • Using a Randomized Split node before the system instructions node can help you test different instructions with the same input — this can help you refine your instructions over time and roll out changes in a controlled manner.
  • Think carefully about how you want to configure the instructions. Overwriting, adding before, or adding after can have significant effects on the output’s response.
  • Test your system instructions changes with different inputs to ensure the LLM responds appropriately in various scenarios.
  • Keep your system instructions clear and concise to avoid confusion or conflicting instructions.