Janitor ai bots...

[Ghost]

I wonder... Does rating a public bot's response affect it's character or is it all in the way it's coded from the get go?

So with that said, is it better to keep a bot private so nobody else can interact with it and affect it's character responses by rating them a certain way?

I wanna hear some of you programmer ai waifucels opinions

 

[Scatius Deletus]

Idk i just rape girls on it

 

[Friezacel]

Pretty sure only the one who created the bot can change its code

 

[aspiecel_11:11]

I think the ratings affect the whole language model (so with the rating you can show the model that the messages in general don't make sense or are out of character), and the character's description is just something secretly added to the prompt.

 

[aspiecel_11:11]

So basically ratings barely affect one specific bot.

 

[Ghost]

So basically ratings barely affect one specific bot.

So... ratings can't turn a submissive bot into femdom or vice versa?

But sometimes I was able to turn submissive to femdom then when I reset the chat and start new convo it's back to submissive?

 

[aspiecel_11:11]

But sometimes I was able to turn submissive to femdom then when I reset the chat and start new convo it's back to submissive?

It's due to chat history that is thrown out with a new chat, with the right (wrong?) messages the original character gets derailed. Your rating can be secretly added to chat history but that's about it.

For ratings to affect the bot across conversations the website would have to store a separate language model for each bot (the simplest ones that kinda work are a few gigabytes so uhh) or figure out how to store the differences more efficiently than whatever these files use.

character.ai, which seems better-funded, wrote half a sentence that the ratings affect a bot (without specifying whether it's within one conversation or across many) with a much larger notice that it might also not, so it might be corporate cope:

Training a Character | Character.AI

book.character.ai