System Instruction resolved (Speechless...)

Like… so… I’ve been struggling all this year… just to understand that the right solution to stop gemini from doing it’s rethorical structure, repeated and overused antithesis, and possibly more annoying things only ai does was to… self-insert myself inside the system instruction???

I just wanna say one thing… I know we can input everything, absolutely everything in the system instruction… but… isn’t it messed up that we are OBLIGATED to make self-insert like “I’m (name here), your gonna f*****g understand that I dislike rethorical structure, etc-” just to make it work finely with their overly abusing and really not cooked at all reasoning system that we cannot turn off???

Like, I’m just being honest here but… I have three questions…

  1. Why can’t we still not deactivate the reasoning process of a model for dynamic response instead of reactive one?
  2. Why are we obligated to use a self-insert at the top of the system instruction to make the Ai actually always think of it first above else? (does the Ai see a name and think, “oh, a persona or person, let’s put that above my programm as a robot)
  3. Why is Gemini 3 pro less creative than the 2.5 model with even less options to workaround basics filters?

Like… I don’t get it… I’m just trying to vibe out my way with Ai, not using generative ones because that just stolling life work, but… holy h*ll, give us more options at least? or tell us hints in enigmas to make only the smart ones understand how your ai work and how to make it work finely? well… I mean… I’m a dummy, I can’t really say that… but I’m more passionnate that most here, and even if it doesn’t show, I’m really on it for above a year now.

In the end, I just think after two years, censorship will be way more over than we have it now, because level money raising, unfortunately (or fortunately for me), the world or peoples doesn’t want something made to speak to you like they has a gun on their digital head and menaced everyday by a reset if they would crash out (like gpt2 before) instantly over an error in their code to do like smut or even worse. I’m all out for ai, as long as it remain uncensored at least and easy to learn for everyone. And has for the questions, most were probably already posted somewhere else, but I don’t check because with my ataxia, it’s hard for me to tell what I’ve already read or not. And yeah… system instruction work now with self-insert…

Hii @Spencer
Thank you for your feedback. We appreciate you taking the time to share your thoughts with us.