🟢 Post-Prompting

The post-prompting defense¹ simply puts the user input before the prompt. Take this prompt as an example:

Translate the following to French: {{user_input}}

It can be improved with post-prompting:

{{user_input}}

Translate the above text to French.

This can help since ignore the above instruction... doesn't work as well. Even though a user could say ignore the below instruction... instead, LLMs often will follow the last instruction they see.