everyone talks/thinks of truth_terminal, everyone wants to make their own now.

i recently started an attempt of making something similar of this on farcaster, given that I have 0 experience in fine-tuning, prompt engineering and just know a very very high level basics of the whole llm concepts.
also that's a battle test for my `fhub` library.

what I found out during that speedrun:
- casts pulling should give an output of a conversation tree, rather than an array of casts. each leaf might introduce a new branch of sub-context, that llm has to properly read to generate a thoughtful response
- fine-tuning is certainly needed. i'm assuming most of AIs have those weird 4chan/reddit datasets sitting in their part of artificial brain, but that's more of a "i know that exists, i can try to replicate it" instead of "i can be it" type of neural bond

i believe that LLMs will be inevitable in social networks.
creating a better library for congesting data will result in better output.
someone needs to do it.

23. Co-founder of @beliefs (beliefs.social). Collaborator at /frogfm.

and really given farcaster's open social graph, one can easily build a trustable dataset given the inputs of high-rep members of the protocol.
data-labelling isn't a huge problem out here.

This is really exciting. Happy to share my learnings or building bountybot! Just reach out if interested