Content pfp
Content
@
0 reply
0 recast
0 reaction
tldr (tim reilly) pfp
tldr (tim reilly)
@tldr
a "Farcaster Lore" packet for Agent training context window.
5 replies
3 recasts
39 reactions
basil pfp
basil
@itsbasil
@peth
1 reply
0 recast
1 reaction
peth.eth pfp
peth.eth
@peth
What do you think the usecase is for an agent trained on Farcaster lore @tldr? For the past few days, I‘ve been thinking about "Farclore" in terms of a curated knowledge graph. General thinking: - Farcaster currently is the best community of positive sum builders in crypto, imo - It already has a richer lore than most people know, and it will only get more complicated - To make it worse, there will be an increasing number of grifters and it will be increasingly difficult to keep track of Farclore would map the evolution of the Farcaster ecosystem, what projects inspired what & which builders participated in what. Their bounty history & attestations as to whether they were good contributors etc. In terms of agents, was only thinking it would make it easy for people to retrieve knowledge directly on the timeline + would be fun to have a Farclore Bard writing songs about these heroes. wdyt @tldr?
3 replies
0 recast
2 reactions
basil pfp
basil
@itsbasil
how would you train this?
1 reply
0 recast
0 reaction
peth.eth pfp
peth.eth
@peth
train, or produce good data in the first place?
1 reply
0 recast
0 reaction
basil pfp
basil
@itsbasil
yeah, feels like hard data to produce?
1 reply
0 recast
1 reaction
peth.eth pfp
peth.eth
@peth
the more time passes, the harder it gets for sure.. most reliable data would be from eg. clanker, seeing who started what project and from bountycaster in seeing who completed bounties for what project but overall, i think all data should be submitted & validated through farcaster, would be something like: - someone makes a claim of how project X inspired project Y & tags the founder, founder then confirms it or doesnt - or the agent notices project Z passed $10m mcap, at which point it asks what projects or people inspired them to start it - if the project has a tipping system, that contributor data could be added in - likewise, agent could at different intervals again tag founders and ask it who are the people that made highest impact contributions, then ask those contributors what they think (to add more people) idk, lot to think about but def think it should be done through farcaster itself at the social graph layer, both the submissions & validation, with the claims in the UI linking to those
2 replies
0 recast
1 reaction
peth.eth pfp
peth.eth
@peth
also wouldnt mind having v1 data just in the hands of people like you & 10 other highly trusted farcaster people, then have each info node validated by the wider community what do you think would be the hardest or most important kinds of data to wrangle?
0 reply
0 recast
0 reaction