Varun Srinivasan
@v
An update on protocol priorities going into 2025. We think a lot of interesting things are going to be built onchain soon. Farcaster should be the programmable and credibly neutral social network that brings users, developers and blockchains together. But what does this mean tangibly?
35 replies
52 recasts
287 reactions
Stephan
@stephancill
1. What impact do you think a warpcast embedded wallet will have on alternative client adoption? 2. Please make it a priority to keep an up to date list of all the notion spec documents floating around. Impossible to keep track
1 reply
0 recast
19 reactions
Varun Srinivasan
@v
I suspect that all onchain apps will move towards embedded wallets. The user experience is just too clunky if its 'install this other app to continue'. The ideal experience is one where the user has a single wallet and grants limited permissions to apps. @gakonst has pilled me on this idea, but its not yet possible and there is a chance that it is impossible to build. The second best experience is separated wallets, but with links that make it easier to transfer funds or do fancier things like spend permissions. @wilsoncusack and the base team are cooking some things here.
3 replies
0 recast
13 reactions
Varun Srinivasan
@v
On docs, yes that sucks today and we will fix.
1 reply
0 recast
2 reactions
casslin.eth
@casslineth
Consider putting docs in a format that’s easier for llm to digest, then can have an llm powered chatbot/search for devs.
1 reply
0 recast
0 reaction
Varun Srinivasan
@v
is there a good example of an llm powered chatbot for docs? i haven't been able to make it not hallucinate responses
3 replies
0 recast
1 reaction
casslin.eth
@casslineth
for example: https://llmstxt.org/ & https://docs.anthropic.com/llms-full.txt can leave application side to community, docs in those format would lower barrier to entry a whole lot for indie devs.
0 reply
0 recast
1 reaction
casslin.eth
@casslineth
I use Claude 3.5 w/ projects ui on their website, i.e. upload lots contents to Claude, confine its role in the system instruction, and then ask the llm, it is much less hallucinated and can actually work well. Think of it as some kind of RAG. So if we do a bit more work on top of a good enough llm model there will be much less hallucination, good enough to be a chat bot.
0 reply
0 recast
1 reaction
Dwayne 'The Jock' Ronson
@dwayne
@v this is what casselin means https://warpcast.com/casslineth/0x3e6ef586
0 reply
0 recast
0 reaction