I am writing various short stories, and I want to use AI to expand my creativity. I can make a prompt, insert my story in the system instructions and then run a prompt to improve it. This fixes spelling errors, style errors and provides instructions to improve the story.
But now I want to write longer stories with a specific set of characters for my stories. So I want to add fine tuning with a structured prompt. However, my sources are stories with just text, not data sets with key/value pairs. Is there a way to fine tune with just texts?
Welcome to the forum. I don’t think fine tuning is the approach that will help you with longer stories; fine tuning might be appropriate for narrow-focus stories, and I have not tested that it would help. There is a cookbook example that illustrates a viable approach to create longer stories with a set of characters and a plot outline developed at the outset here - Google Colab
Hope that helps.
Well, it’s not just for writing longer stories. It’s for fact-checking my longer stories to make sure there aren’t conflicts with the prior story. So, for example, if I declare a birth mark on the left cheek of one character and then later, in another story, I put it on the right cheek, then the AI would notice it and suggest a fix.
I don’t want the AI to create the story. It just needs to find the flaws.
Hi @Katje
I think for your use case clear and specific prompts will do the work. Gemini 1.5 Pro already comes with a 2-million-token context window, so that will fit your story and point out any flaw as per instruction given.