#Need help to train an article writer

4 messages · Page 1 of 1 (latest)

dim furnace
#

I have an archive of thousands of my own articles that are common. It’s the same industry every time, same broad topic, same format/approach to each, similar length… a journalistic article, pivoted off an interview, to present the speaker’s insights, interspersed with my analysis and context.

Going forward, I am keen to investigate using AI to do this - I guess, by fine-tuning text-davinci-003 or the later ChatGPT API.

How viable is this, and what do I need to think about?

Prompting

  • I see that fine-tuning is actually a process of providing a volume of prompt-completion pairs. I assume the completions I provide should be examples of my finished articles (?). But what about the prompts... ?
  • Transcripts of the initial interviews are available in very many cases. Should the prompt comprise only of the corresponding starter transcript... ?
  • Or should they each be topped by some writing instructions (for what to do with the material)?

Guidelines

My writing process, and that of the AI, should follow particular norms around:

  • Headline style
  • Presentation of interviewee quotes
  • Attribution style
  • Tonality
  • Objectivity
  • Tense
  • Pattern of cross-head use
  • It should always preserve speaker quotes, when used as direct quotes, in quoted text.

When the speaker is quoted outside of a direct quote, the attribution should include novel paraphrasing, never repeat, never support or endorse etc. Also, Idon't want it just making stuff up.

Lately, I have been testing this out using ChatGPT and text-davinci-003, by issuing instructions that I believe describe what I do manually. But the results have been mixed, and the prompts have been long. I assume fine-tuning would give it something more specific to mimic - me.

hazy crystal
#

I found fine-tuning to be really tough and didnt succeed on that at all, but putting relevant pieces of your texts into the prompt that you previously look up via similarity search on embedding vectors works reaaly good: https://dagster.io/blog/chatgpt-langchain Special requirements to the ouput can be adressed via prompt engeneering. Maybe that helps 🙂

#

its referred to as few-shot learning if I got it right

dim fern
#

Fine tuning is barely helpful in my experience. Your best bet is to just find an article you have published and say "using the style of the writer at: LINK"