#The characters I m trying to get it to

11 messages · Page 1 of 1 (latest)

quasi talon
#

Would you share your prompt and/or screenshots of the outputs, and discuss what you want that you are and are not getting successfully?

I and others are likely to enjoy trying to help you get what you want.

stray star
#

Particularly, I express that I'm new to the universe and no one knows me so it's having the characters say, "I don't know you. Who are you?" which is good but then in the next scene they're basically divulging all their closest secrets and feelings

#

There's no Human progression which I can kind of understand

quasi talon
#

Sweet. Please share your prompt.

#

I bet I can experiment with that and get the characters interacting with yours to be more wary, but it HELPS HUGELY to see what you're already asking for

quasi talon
haughty needle
#

The character that ChatGPT is 'playing' is a character which, by default, means to provide information. If it is given no other details about this character, it will behave according to its default. Specify that this cartoon character must behave in a certain manner, and it will do so. Just imagine what an alien would think of a chair, had it never seen one before and had it never thought about what that object is most commonly used for. That is basically the way AI perceives objects. It knows nothing, it is given training data and intelligently constructs a response to your input based on that data. It has no real understanding, and the current version of ChatGPT does not 'learn' from your responses.

You have to be veeery specific, as if you are a customer at the world's fanciest restaurant. If you want the steak to be cut into Gordon-Ramsay-shaped bits and you would like every third piece to be twice as big as the other bits, then you will have to say so. The chef will try its best to do what you ask, so be careful about how you put it. It can read context, but it will assume which contextual implication to emphasise in its response, which it will determine based on its training data as well as the context of your conversation.

I once asked something in here, then I thought, I could literally write this problem into ChatGPT input box and then tweak its response by saying "Make it sound less like a monologue, make the character secretive." then that would bring it nearer to what I want, and then seeing what it produced, I would say "Yeah, like that, but without the character sounding shady - they simply don't divulge everything about themselves from the get-go, the character is having a normal conversation with a stranger."

stray star
#

I've been supplying chatgpt with 10-20 quotes from these characters and having it build a personality profile and they seem to get it more or less correct but there are some issues.

#

For example, they have the characters speak much more than they would, be more open than they would. Not be cautious or guarded enough, etc.

haughty needle
#

yeah even when you put it clearly and plainly, the model is too self-absorbed like its politically correct creators