Varun Srinivasan pfp
Varun Srinivasan
@v
What makes an account "inauthentic"? We're considering a new labelling system for inauthentic accounts. These accounts, often incorrectly called bots, will be hidden from like counts and other metrics in Warpcast. Inauthentic accounts give real people the ick because they're clearly trying to get value for their account at the expense of everyone else on the network.
46 replies
36 recasts
235 reactions
Varun Srinivasan pfp
Varun Srinivasan
@v
We're looking for input from the community on behaviors that make an account inauthentic. It's important that the behavior is specific and deterministic, like: - follows 100+ users in 5 minutes - has the same pfp as 1000 other accounts - tags 100+ users asking them to follow them back. Please avoid suggestions that are subjective or hard to evaluate. - posts pictures copied from the internet (hard to evaluate) - posts slop or uninteresting content. (subjective)
13 replies
3 recasts
33 reactions
Varun Srinivasan pfp
Varun Srinivasan
@v
We will use these rules to guide our machine learning model to find inauthentic accounts quickly. The actual model may look at many more facts than just these rules, but they will be a good starting point.
1 reply
0 recast
23 reactions
Stephan pfp
Stephan
@stephancill
@geoffgolberg has identified a bunch of patterns in clusters of inauthentic accounts
0 reply
0 recast
3 reactions
Tony D’Addeo pfp
Tony D’Addeo
@deodad
- frequently quote casts users (10+ per day) and rarely gets any engagement from OP - frequently replies (25+ per day) to users and rarely gets engagement from OP
0 reply
0 recast
0 reaction
Yinks pfp
Yinks
@yinks
Follows 100+ users in 5 mins? But there are people that join through someone that referred them so might end up following everyone on their starter’s pack. Why would they have to suffer for that?
0 reply
0 recast
1 reaction
Bitfloorsghost pfp
Bitfloorsghost
@bitfloorsghost.eth
Honestly the biggest issue to me is some people I truthfully don't know if they are a bot or a real person because their canned ChatGPT responses
0 reply
0 recast
1 reaction
DV pfp
DV
@degenveteran.eth
Does the follows 100+ users put into account the "packs"? So if they do click follow all via those it doesn't trigger inauthenticity
0 reply
0 recast
1 reaction
Said pfp
Said
@said116dao
Why do you put so much emphasis on content? It's a social network, I can't just come in and write some nonsense? Just because I want to. Do I need to write only cool content?
0 reply
0 recast
0 reaction
Patricia Lee pfp
Patricia Lee
@patriciaxlee.eth
- templated engagement (same reply across multiple casts) - link or image only casting with no commentary - churning on unreciprocated follows repeatedly in short time spans
0 reply
0 recast
0 reaction
Camila 🫂 pfp
Camila 🫂
@camilags
Ask for tips. Unfortunately there are many accounts that make comments just asking for tips.
0 reply
0 recast
0 reaction
nuconomy.⌐◨-◨ pfp
nuconomy.⌐◨-◨
@nuconomy.eth
The follow flag could get triggered for a user bootstrapping with lists.
0 reply
0 recast
0 reaction
Mesut pfp
Mesut
@mesut
integration with reputation protocols, Orb, user verification steps, blue mark... You can simply call the rest inauthentic. You have to make every account authentic by default if you want to solve the problem.
0 reply
0 recast
0 reaction
Arjan | That Poetry Guy pfp
Arjan | That Poetry Guy
@arjantupan
Yes, anything like follow for follow, in casts or bio, is suspect. Also, accounts that send majority of DMs to accounts they don't follow. Supporting eachother is fine, but if say 90% of casts & replies feed are 3 or less words, it is suspect. Users that reply the same reply to different casts in a short time without other casts or replies inbetween.
0 reply
0 recast
0 reaction