Personalities: You no longer need to jailbreak Google's Gemini (And other AI models)

I'd like to showcase my webapp which I've made with pure Javascript.

Yes, I know what you're thinking: A SPA with Javascript, nevermind Typescript, and no frameworks? How do you handle state? How do you handle navigation?

Honestly, this was merely a learning exercise to get a good grasp on JS. I promise I don't usually write software this way. ૮ ˶ᵔ ᵕ ᵔ˶ ა

Anyways...

Zodiac -Lightweight, and featureful

Honestly, the greatest feature of this frontend is Personalities.

Personalities are a contained set of behaviors that the model MUST abide by.

This method of prompt injection is even more effective than system prompts at swaying the model's generations towards your purported goals. The result is generations that are quasi-uncensored, no JAILBREAK NEEDED!!

For now, the way to set up personalities is through specifying a prompt. This is how it works:

• The prompt is injected in a premade conversation between the model and the user.
• We then inject a response in the name of the model, in which the model agrees to abide by the prompt.
• Every subsequent generation tends to not contradict this exchange.

This method has the potential to be at least twice as effective the more messages you introduce in the chain. These messages serve as a bias for subsequent generations.

In upcoming updates to this tool (coming in a few days), you will be able to add tone examples, and give the personalities access to the internet, or even your own code.

You want to try it? Check here for a live demo, or self-host it yourself. You will need an API key (free of charge, here: https://makersuite.google.com/app/apikey)