It’s a potential solution for a clever developer who has an interest in this twitter topic.
However, it does sound like, if you do not mind me saying so @Arthur, is that you may not want to put the time into developing this code yourself to accomplish the task you wish to accomplish. Finding an algorithm to generate prompts from tweets is not “rocket science”, if you think about it creatively. It’s not something I am interesting in, but it sounds like it interests you, so go for it!
Thanks for inviting me to this conversation. I think most people that I talked to haven’t find a lot of success using the fine tuning API. But if @arthur and @jseveno.piltant want to try using prompt engineering, it’s probably easier to setup and troubleshoot.
The way I imagine that can work is for every new tweet you want to create, you give it a couple samples of your existing tweet and telling GPT that is the style you want.
You can use the embedding API to search for similar tweets to find the top n number of matching tweets.
The drawback is that this will cost more because it will use up more tokens.
To set this up, you can write custom code or the spreadsheet I have here Google Spreadsheet + GPT3
and store all your tweets in the google sheet to generate new tweets.