Using Facebook data for A.I work

[polecat]

Download your Facebook posts archive (in settings and privacy > export your data somewhere)
Drop the posts json file into Claude or ChatGPT.
Tell it to analyse all the posts to learn how to capture the tone of voice, personality, humor, writing style and insights so it can clone you.
Be persistant and tell it to look at more posts a few times to get the whole thing.
Ask it to tell you what it learned.
Tell it to write a detailed and comprehensive prompt template in which you can ask it to write about any given topic.
Let it generate example posts until you’re happy with the outcome, then collect a few dozen of those, feed them back into the chat and tell the LLM to refine the main prompt to ensure consistency.

One of my friends told me about this today plan to try it out a lot.

 

[Zwielicht]

Yeah, I can see that working. I don’t have a Facebook account, but I trained an LLM to mimic the way I write for different tasks before, and it did well for the most part.

 

[polecat]

Yeah, I can see that working. I don’t have a Facebook account, but I trained an LLM to mimic the way I write for different tasks before, and it did well for the most part.

Makes sense the guy who told me got a pretty unique way to get popular on social media

 

[t2van]

Download your Facebook posts archive (in settings and privacy > export your data somewhere)
Drop the posts json file into Claude or ChatGPT.
Tell it to analyse all the posts to learn how to capture the tone of voice, personality, humor, writing style and insights so it can clone you.
Be persistant and tell it to look at more posts a few times to get the whole thing.
Ask it to tell you what it learned.
Tell it to write a detailed and comprehensive prompt template in which you can ask it to write about any given topic.
Let it generate example posts until you’re happy with the outcome, then collect a few dozen of those, feed them back into the chat and tell the LLM to refine the main prompt to ensure consistency.

One of my friends told me about this today plan to try it out a lot.

You could cut out lots of the trial and error if you ask it to deep dive research your posts and then set a specific learning task for that research.

Re feed it the research to refine the prompt and outcome of whatever you want it do.

Cuts down on the multiple refing with a prompt as it will never learn, that's where research and creating project instruction and providing project assets pays off