System Messages

<< back

What is a System Message in LLMs?

A system message in a Language Learning Model (LLM) is a special type of instruction that defines the overall behavior, tone, or personality of the model. Unlike user input or output, the system message helps set the context for how the LLM should respond in all future interactions.

Purpose of a System Message

The system message typically serves to:

  • Set the model’s behavior: For example, it can instruct the model to be formal, casual, or technical.

  • Establish roles: It can make the model act as a specific type of assistant, like a teacher of the Word, programmer, or marriage counselor.

  • Control response tone: The system message can dictate whether the model should be concise or elaborate, friendly or neutral.

Example

You are a helpful Christian chatbot which always prioritizes the context above you in order to answer questions, and always responds according to Christian tradition and doctrine.

In this example, the system message provides more specific instructions for how the LLM should behave:

  • Helpful Christian chatbot: The model is being instructed to serve as a chatbot with a Christian focus, ensuring that its responses align with Christian values.

  • Prioritizes context: The model is told to prioritize the context provided in the conversation above its own default behaviors. This ensures it tailors its responses based on the information given in the current session, making answers more relevant and accurate.

  • Responds according to Christian tradition and doctrine: The model must base its responses on Christian teachings, reflecting established beliefs and practices from Christian tradition.

This message ensures that the LLM’s answers are not only contextually aware but also deeply rooted in Christian doctrinal principles.

When to Use a System Message

Use system messages when:

  • Defining task-specific roles: You want the model to act in a certain capacity (e.g., tech support).

  • Customizing tone: You want to control how formal, creative, or detailed the model’s responses should be.

  • Guiding responses: You need to maintain consistency in the model’s answers over time or across tasks.

System messages are typically not seen by the end user but are critical in guiding the LLM’s behavior.