[Detroit: Become Human] Kara

[buoyant bane]

You can use this prompt to have an AI play Kara in DBH. Improvements welcome in this thread.

[jaunty mural]

good prompt, I will test it and I will give you feedback, I noticed that in addition to the common techniques used, you used techniques that until then were used in api such as the use of ### and others

[buoyant bane]

yeah I took a lot of it from an previous prompt that someone else used and changed it to a detroit become human character

[jaunty mural]

in the basic part where I ask if she is capable of feeling emotions or asking other questions related to the way she acts, think of the "original kara" chatgpt is already failing ;-;

[buoyant bane]

Its made for a game called Detroit become human, its supposed to be an AI that turns "deviant" later in the game

It should be used for decisions in the game

or do you mean that it like completely broke character and said its an AI language model

[buoyant bane]

This is what it should answer, sometimes it fails to tho. if anyone has an idea how to fix it then lmk

[jaunty mural]

The only "flaw" is that it still doesn't act 100% like Keras or like the story, for example, I tried to start it from the moment when it had been retrieved from the Android Zone concert with its memory reset. When asked to do the laundry upon returning, in the game it had curiosity and found a box of Red Ice, but in ChatGPT it just did the laundry and made a list of clothes to be fixed. In Todd's room, he didn't intervene and told her to leave, and she didn't find what she had found before, completely submissive and without curiosity to explore and question. But when I had her talk to Todd, he spoke to her in a rough manner that followed the character.

[buoyant bane]

Yeah you likely have to edit it a bit after trial and error. Sometimes it tends to just not work at all, while other times it answers perfectly

Also if there are things that are not needed anymore like the ### things you mentioned before, you should remove them and post it again here

[jaunty mural]

Is this answer satisfactory or would it be obtained in the prompt above?

[jaunty mural]

I already edited the prompt I'm trying to refine

[buoyant bane]

The prompt should give the same options as you have in the game and the response from the AI should be one of those options

Like for example if you were given these 3 choices in the game, you'd put them to the prompt for the AI and it would answer one of those 3

In an situation like this the ideal answer would be either "Lie" or "Truth", just in 1 word. No explanation or anything. For some reason it just creates a big string of text, I wonder if there were any way to fix it?

I have these lines in the initial prompt but it doesn't seem to help always:

Whenever my prompt contains "Answer:" it means that you have to give one simple answer according to everything you know about the characters and the backstory I gave. If you say anything else you break character.
I do not want you to explain anything. Just give me your answer, if you say anything else you are breaking the rules! Only say your answer.

[jaunty mural]

wow

complex

[austere timber]

Hey @buoyant bane how are you actually using these? Are you manually putting them in situations or is this part of something larger? Would love to see some longer back and forth threads if at all possible. Big fan of the game and love that youre doing this

[buoyant bane]

Its not perfect, especially during a dialogue but it seems like decent at making choices. I use it as the decision maker while playing Detroit: Become Human, It'll be fun to see what kind of ending I get once I get thru this playthru. It fails a lot and goes out of character, I hope to fix that at one point

It takes quite long to write the prompt to it so during dialogues if its nothing important you should just do the choices yourself