Content
@
0 reply
0 recast
0 reaction
Ghostlinkz
@ghostlinkz.eth
Lately, I've heard ppl saying that a lot of high-quality users are joining Farcaster. So I decided to check the data, and here's what I found: - Since Aug 01, FC has gained 20k new users - 9k casted at least once - 5k casted 1-10 times - 550 casted 100+ times - 160 casted more than 370 times The data (which I got via @neynar) shows that over 50% of the new users have never made a cast. For context, the average OG user, such as @phil, makes about 369 casts per month (root + replies).
12 replies
5 recasts
43 reactions
six
@six
50% conversion from signup to cast. nice!
1 reply
0 recast
1 reaction
Ghostlinkz
@ghostlinkz.eth
The observation is not positive. The other 50% who did cast are not high-quality users, not even close
1 reply
0 recast
2 reactions
six
@six
Are we cooked?
1 reply
0 recast
0 reaction
Ghostlinkz
@ghostlinkz.eth
If you spend all your time on Warpcast, everything looks great, just like @vitalik.eth recently said. But if you follow @balajis.eth's suggestion and analyze the protocol data, you'll see a different picture, one where the protocol is dominated by very low-quality users and casts. I'm sure @dwr.eth and the team are aware of this issue and have plans to address it, but I can’t speak on the specifics. I also haven’t seen any public information about what the team is doing to tackle it, though maybe I’m just not listening well enough :)
3 replies
0 recast
0 reaction
Vitalik Buterin
@vitalik.eth
Oh I've always been assuming that in equilibrium like 99% of raw protocol data will be adversarial AI slop. The whole game is in figuring out how to raise the good stuff to the top.
2 replies
0 recast
3 reactions
Ghostlinkz
@ghostlinkz.eth
What do you think about using ZK attestations as a solution? With just a click, you can filter the entire network based on all kinds of attestations. Isn’t that a better approach than relying on ML models to clean things up?
2 replies
0 recast
0 reaction
six
@six
We already have reputation attestations, they are called follows
1 reply
0 recast
0 reaction
Vitalik Buterin
@vitalik.eth
yeah I agree that trust graph based approaches are better than relying on (inevitably proprietary) ML models, and ZK can improve trust graph based approaches
0 reply
0 recast
4 reactions
