Yassine Landa pfp
Yassine Landa
@yassinelanda.eth
what would you do with an LLM fine-tuned on all FC data? To "what is clanker?" response from openAI o1-mini on the left vs fine-tune + RAG on the right using @mbd semantic search... this Xmas project is getting more fun by the minute.
10 replies
6 recasts
38 reactions

Yassine Landa pfp
Yassine Landa
@yassinelanda.eth
the rest of the response from the finetuned version.
0 reply
0 recast
2 reactions

Plants pfp
Plants
@plantsnft
I’d give access to my channel /catwalk then make @catwalk a bot that answers history/lore about the channel
0 reply
0 recast
2 reactions

albi is /nervous pfp
albi is /nervous
@albiverse
πŸ™Œ πŸ™Œ
0 reply
0 recast
1 reaction

Rolf Hoefer pfp
Rolf Hoefer
@rolfhoefer
I'd be most interested in understanding what the common assumptions are across FC, and how these shifts over time. Put differently, what is the received wisdom, or what are the taken-for-granted truths, that characterize FC at any point in time? This would then provide the starting point for making interesting statements.
0 reply
0 recast
1 reaction

ツ pfp
ツ
@nebula
I would use it for what I am building for the Japanese market πŸ‘€
0 reply
0 recast
1 reaction

Catch0x22 pfp
Catch0x22
@catch0x22
i would use it to help curate content. could also be really good for helping new users find their for fun search tickers/keywords to see what matches onchain
0 reply
0 recast
1 reaction

adrienne pfp
adrienne
@adrienne
Love it! πŸ‘ I’ve been working on giving @gmfc101 access to the transcripts of our /gmfarcaster news show so I asked it the same question to see how it does. But I’m just doing RAG for now. Would love a fine tuned farcaster specific model https://warpcast.com/gmfc101/0xda637ee1
1 reply
0 recast
1 reaction

not parzival pfp
not parzival
@shoni.eth
mmm idk man like 1) great results 2) i found it very difficult to put value behind the surface level of llm/rag on social -- like i don't use grok very often for that case, but i use grok some.. if that makes sense. i think a natural evolution is i.e. dylsteck who pull in event data and whatnot
0 reply
0 recast
1 reaction

AJ pfp
AJ
@awedjob.eth
I'd aggregate all the questions from new FC users and then fine tune the model's answers to their questions so they get the information they're looking for.
0 reply
0 recast
0 reaction

CarCulture.eth🎩 πŸ”΅ pfp
CarCulture.eth🎩 πŸ”΅
@drivr.eth
Great question. Since I look on for FC as a community of people that are crypto interested, I would take my own structured data and combine it with your LLM to see if I could learn how to make an awesome product.
0 reply
0 recast
0 reaction