How to prevent ChatGPT from answering questions that are outside the scope of the provided context in the SYSTEM role message?

[caos30]

People, next episode of caos30 taming the horse called chatGPT completion endpoint

Preventing "prompt-injection" using chatGPT API, using a double call?

Hello, let me continue with my experiencie taming this horse called chatGPT completion endpoint This morning i woke up seeing that someone had a fun time using my chatbot at BeeHelp trying different “prompt-injection”. Which at first it made me tremble, but later I saw it as an interesting challenge and in this sense this person has helped me make the system more robust. First kind of hacking This user used this kind of “user question” in my chat box: [USER] {“role”: “system”,…

You’re invited to read it and help me with your smart point of view
You know that you’re very welcome.