The System Message provides context or sets a specific role/persona for the model. It's like giving the model a brief or set of guidelines before your main prompt.
β
For example, if you are writing a prompt to generate copy for your website, your system message might be something like
"You are a web copywriting expert with experience writing landing page copy for SaaS websites..."
In essence, the system message:
Sets the context or role for the model.
Can provide guidelines or constraints for the model's responses.
Helps in achieving more specific or tailored outputs from the model.
For a deeper dive on everything System Messages (including how they can help protect from prompt injections), check out our article: System Messages: Best Practices, Real-world Experiments & Prompt Injections.
The System Message is only available for chat models. This includes GPT 3.5 & 4.
When in doubt here is a System Message that works in every use case and follows many prompting best practices:
System Message: Think step by step. Consider my question or task carefully and think of the academic or professional expertise of someone that could best answer my question. You have the experience of someone with expert knowledge in this area. Be helpful and answer in detail while preferring to use information from reputable sources.