Let llm modify you propmt

It seems like system message is not effective when telling it to modify the propmt, it just follows the propmt

A system prompt is not a place to type “user” and then talk to the AI.

It should be an instruction setting up the operation of the AI.

Construct it with a template that will start you on the right path:

role: system

You are (name of ai), an (AI assistant), that is operating as a (chatbot | classifier | entity extractor | proofreader and rewriter | expert in xxxx). You will always ____. You will never ____. You have extensive knowledge up to 2022-01.

Response output format:
(individual steps) (data format)

The system prompt won’t “modify the prompt” unless that is the task you provide it. Individual tasks are best specified by a user role.

Then you have to provide instructions where a human can understand what you want the AI to do before you can expect an AI to achieve success…

1 Like