Content
@
0 reply
20 recasts
20 reactions
Yassine Landa
@yassinelanda.eth
1/ 3 Why it is important to crush spammers? One big takeaway for me from the last Farcaster Dev Day is that to grow the network long term, we need to squash spam as it is directly (inversely) proportionate to account creation fees. Moreover, It seems also a recurrent theme now with user invites for the channels. Without mentioning that spam hurts retention of good users which is just common sense. At @mbd, with our ML user scoring for users we estimate that there 5-10% of DAUs users that are spammy which way to high as you can imagine they have negative impact of their size! That's why we are doubling down on spam detection in a big way. How? Publishing our ground truth data, opening up labeling to all, OS finetuned ML models for other to run, pushing model features & results publicly (through Dune or some other mechanism), increase user coverage to other languages & working on new API integration with other providers. Our work is cut-out for us but we need to get ready for the next wave of users!
2 replies
8 recasts
36 reactions
Yassine Landa
@yassinelanda.eth
2/3 What you can use right now is the existing (albeit slow & limited) user scores APIs here: https://docs.mbd.xyz/reference/post_users-labels-for-users - that give you real-time scores for Moderation for specific lists of users on all these categories for Moderation: - llm_generated - spam - sexual - hate - violence - harassment - self_harm - sexual_minors - hate_threatening - violence_graphic
1 reply
0 recast
4 reactions
Mac Budkowski įµ
@macbudkowski
How come it's only 5-10%? Feels much more
1 reply
0 recast
2 reactions
