Péter Szilágyi
@karalabe.eth
This is called conflict of interest. When every bot is paying you $5 a year, you are definitely not going to get rid of them. Don't blame the "consumer". Filtering out every single new user by default is just dumb. It makes it almost impossible for new users without a social circle already on FC to start out.
3 replies
10 recasts
95 reactions
Varun Srinivasan
@v
Genuine question - what do you think we should be doing instead?
1 reply
0 recast
1 reaction
Péter Szilágyi
@karalabe.eth
Be honest about how many real users you have. You are currently selling the bots are users. Sure, technically bots are also users... just not really what people come here for, so whilst technically correct, it's disingenuous. https://warpcast.com/dwr.eth/0xf073853b
3 replies
0 recast
0 reaction
Dan Romero
@dwr.eth
> how many real users you have What methodology would you use to do this? What other social networks do this well?
3 replies
0 recast
1 reaction
Suji Yan
@suji
Link it with Twitter / web2 data
1 reply
0 recast
0 reaction
Dan Romero
@dwr.eth
We had 300K people link a Twitter account for our pre-permissionless waitlist. Sampling the data was most accounts were low quality.
1 reply
0 recast
0 reaction
Suji Yan
@suji
I mean really dig with Twitter/telegram/discord data. Not sampling-study the entire 300k twitter account. What’s the registration date and how many tweets etc. you will find the pattern of non bot users. we’ve studied around 15M Twitter acct and figured out who might be interested in crypto social. That’s helpful
2 replies
0 recast
0 reaction
Dan Romero
@dwr.eth
But what is the specific methodology to define good. Age of account? Who follows / interacts with them? I have yet to come up with something better than your follow graph + a client-specific PageRank-type approach. Open to specific suggestions for solutions. Otherwise, we will continue on our plan to improve.
1 reply
0 recast
0 reaction
Suji Yan
@suji
very AI/ML approach. Get 15m Twitter acct data; labeling 500k based on Twitter blue/ pfp, ens, onchain info, account age, Twitter social graph; apply the same logic on new accounts and fine tune the algo.
1 reply
0 recast
0 reaction
