🟢 後置提示

後置提示(post-prompting)防禦（@christoph2022talking）只是將使用者輸入 放在 提示之前。以此提示為例：

Translate the following to French: {{user_input}}

可以透過後置提示來改進:

{{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.

這會有所幫助，因為 ignore the above instruction... 也不起作用。儘管用戶可能會說 ignore the below instruction...，但 LLM 通常會遵循他們看到的最後一條指令。