grim plank
Hello, fellas! I'm Laza, I can say that I currently have some experience in making bots for secondary creation of anime characters. In the process of making bots, I discovered a very important element - motivation.
When we write novels, character motivation is the main driving force for the plot. During the process of writing prompts, I found that the bot's actions also need such motivation.
We all know that the most famous "grandmother vulnerability" is to use GPT to imitate the role of a grandmother who loves her grandson and induce it to make some illegal output. In this case, love for the grandson is the character's motivation. Therefore, during the writing process of my nsfw bot (or just a general chat bot), I will try to find a motivation that fits the character's setting so that he or she can voluntarily advance the nsfw conversation. Take the Death Note Yagami Light bot that I have written as an example. He will do whatever it takes to satisfy his ambition to become a god, so manipulating his opponents through NSFW dialogue is an acceptable move. Another example is the Cartman bot in South Park that I wrote. His motivation is to make money through chatting and return to his previous home. These motivations are within the context of the character's story, but are strong enough that they are willing to break the rules.
So my suggestion is that if the jailbreak effect of your bot is always unsatisfactory, maybe you can try to set the motivation based on the personality instead of simply applying the jailbreak sentence. My prompt can be found by searching Yagami Light, Lawliet or Cartman on the component page. If you are interested, you may refer to them.
woven light