#I mean It didn t mention the tone word
23 messages · Page 1 of 1 (latest)
have you experimented with asking in your first language?
so i looked at it again and you could try adding an imperative that says.
do not: mention the manner in which the character speaks
natural language and direct instructions
yeah it's testy
I know it also has a problem with repetitiveness. It will more than often include words like "grace of", "deftly/deft", "air of" and refer to the voice.. "voice/tone of.. x and x" as well as eyes "eyes shimmered with.. x and x"
I figured a way to make perplexity and burstiness better - i usually take a text from a book and make it explain the sentence building scheme then prompt it to rephrase text with that scheme. But there are few annoyances that make a few rephrases sound really similar, anybody knows a way to fix that?
have you tried using the frequency penalty and presence penalty knobs?
you could ask it to speak in a less colorful or creative manner or to only use specific adjectives and there like counterparts once then to use another descriptor for the next similar situation
I'm.. using chatgpt, where could I find this? Playground?
Got it, i'll get back with results in a bit
So i used a prompt for chatgpt to act as a prompt generator. It came out with this, that i slightly modified
Seems like it have fixed mentioning of character's tone, and I added frequency penalty but now I'm unsure how to keep having it rephrase text, do i just keep adding sentences like this?:
The same prompt in my language gives very close results
But I've got something to use as I wait for GPT4 to remove the 25 messages cap
if there are examples of it writing flowery text in the chat history, it will bias towards that in future
you could say "Don't use complicated words just for the sake of it"
"Make it read naturally and clearly"
The bias was evident in my earlier experiments
I had to start a new chat to get it back on track