Jailbreaking is an inherent problem with LLMs that can never be solved. Any safeguard has to be less capable than the LLM it protects, else you can just target that. So there will always be a way to communicate with the LLM in a way that bypasses the safeguard.

It’s like trying to sanitise user input from SQL injections, except the database speaks every form of communication documented by humanity.

All this is to say, I’m glad I’m not responsible for any of these systems.

Oh yeah. Aight, I put on my robe and wizard hat.

One thing that gets me about AI chat agents is the idea of attack surface. If you have a clearly defined protocol you can curtail most of the possible attacks by narrowing things, only accepting well formed requests, and validating both on the user end and then on the server end before processing anything. An LLM is inherently wide in attack surface given the way it is structured. It can take a prompt which can be any set of characters connected together into tokens. These tokens can't easily be filtered for intent or goal and yet they can get the LLM to drop other rules or restrictions because they are just other prompts.

A simple coded padlock is not very secure, but a door with no walls is less secure.

So prompts that define THE GenAI personality?

But can it be my waifu?

Did just amazon and more using know names to fire them of knowledge. Because Rufus is the tool to burn CDs and record operative systems

Is there a place that collects which prompts can be used for these things in an up to date way?