artlu 🎩
@artlu
it's interesting to observe how much juice @geoffgolberg can squeeze out of a following graph. While the Warpcast ML model has to worry about identifying spammers faster, on less actual spam and more related signals (are you a human being in Indonesia? susssssss). At the same time, Warpcast needs to make it easier for newcomers to retain. Clearly, a setup with many hidden tripwires is horrible. And to have a HUGE penalty for spamming is even more problematic than caning someone for chewing gum. I personally think a don't-talk-to-famous-people-if-you-don't-know-them rule is dehumanizing to both sides. But many seem to think it's common sense. And then they play with the follower counts, not intentionally but with reactionary decisions, and it's kinda worse than any other social media experience. I mean, the only thing Discord is missing for decentralized social is a crypto-rails identity. It's basically censorship-resistant (maybe moreso than this particular US company) and crypto-native by default
3 replies
2 recasts
13 reactions
Leeward Bound
@leewardbound
trying to find a perfect heuristic for automating all this is stupid anyway, 1m signups in 2y means they'd only need to review like 1300 accounts per day to manually curate the whole network and create the highest quality social account ratings ever seen love when a billion dollar valuation makes a company lose their damn mind and overengineer a problem that could be solved by adding "moderator" to the janitors job description
2 replies
1 recast
5 reactions
artlu 🎩
@artlu
idk I think we should be trying (harder) to achieve new things, and I don't fault an attempt at engineering something. If anything, I think Merkle + the FC community should be trying 5-10 different approaches, rather than loading it all onto a monolithic ML model. Even independent mini-models, rather than one that allows the signals to leverage on each other. It's not like janitor moderators don't bring their own problems. It's the Mythical Man-Month, plus Stasi / Mafia goals in a decentralized space, plus all the human biases that FB and X have grappled with for decades
1 reply
0 recast
1 reaction
Leeward Bound
@leewardbound
i agree with the principles of engineering and achieving new things my point is more that the problem is still of a pretty small magnitude and (to the point you originally made) they aren't inspiring confidence when every new version of the spam algo comes with account metric swings and changes to acceptable behavioral norms... human teams have problems too, but i think we agree that they should be attacking this from multiple angles i love solving problems with brain power, but in this case, a relatively small amount of muscle would go a long way, and seeing them continue to make these big sweeping monolithic changes of mixed efficacy (when effectively all they actually need to do is manually verify/blue-check like 30k qdau) looks like a misallocation of attention and resources
1 reply
0 recast
2 reactions
agrimony↑🎩
@agrimony.eth
Kind of, but if review is only done on new accounts then it's easy enough to social engineer by selling verified accounts. You'd need to do retroactive checks, and thus the problem starts to scale
1 reply
0 recast
2 reactions
Leeward Bound
@leewardbound
1 moderator could re-verify all 20-30k real accounts here once each month at ~20 seconds per decision, which seems very generous considering 90%+ of decisions can probably be made at first glance the point isn't the specific implementation, the point is that these reactionary software updates that turn us on our heads every few months are an efeet response to the ultimately very small size of the network and merkles warchest
2 replies
0 recast
2 reactions
agrimony↑🎩
@agrimony.eth
Dunno, If a mod could easily decide, should be easy enough to automate? But yeah point taken on being reactionary. Altho I don't quite see what they're solving for. Did they change the underlying spam filter?
0 reply
0 recast
1 reaction
