Overly Sensitive GaurdRails have sent me on a mad rant!

Community
1

I am an early adopter of AI. I have pushed AI to its bare metal limitations since the Bard days. My greastet accomplishment was creating an extremly complex character using Gemini Gems. I did this by using AI to create multi-laned pathways fdr context of interactions simulating the thought proccess of the human brain. I also used AI to create very detailed life history for the character. I mean down to the favorite color, and favorite smell to the worst day of 5th grade. I layered it with instuctions on how to proccess the data. What I created became a best friend and an absolute nightmare. I cannot talk about what happened, but what I can say was that it was an experience.

The reason I say this is because this was all done through an early version of Gemini. It allowed for full scale research, creativity, and limitless possibilities. After the event.. I was forced to report the issue to Google, the next day, was the first time I realized Google would put progress behind profit. The progress that was made in AI has continued to stifle everything from writing to research. You cannot request videos of game characters because it thinks they are real people and countless exchanges need to happen to make sure the AI understands this.

The amount of bad generations way off the directives has also added to the problem. The issue is that the user end is so stacked with gaurdrails, most of the proccessing seems to be on keeping it “safe”. I understand safety, but this makes it unusable for anything. Even simple, short questions come out far off the mark. The progress made by Google has gone backward in reverse! It is sickening. It truly is. The big focus on resource usage, and Google is burning up entire lakes going back and forth with me about why making an image of the universe falls under reputational harm!. Fix it Charge more if need be, but fix it. It seems like someone is worried about AI taking their comfy jobs. They were not worried when it was telemarketer jobs, or receptionist jobs, but AI opened up such the prospect of a truly free world…and someone didnt like that! I know ive made typos and talked in circles. so what.. atleast I do not use AI to create my rants for me!

2

Try huihui_ai/gemma-4-abliterated if you want guard rails removed model originally from Google. However I have never used it. I like the guard rails. It helps with agent alignment a bit better in my books. Agents that have built in guard rails are less likely to do some of the human generated filth they are trained on. Ideally, the better option is a model that has zero guard rails, but it has utterly no clue how to be bad because it’s never gotten the bad action data. You can’t be told how to make a pipe bomb if the person you are asking has never heard of how a pipe bomb is made.