The problem is “spam” and not “bots”. A lot of these accounts are human-operated even though they post copy pasta’d or llm generated content, and can pass proof of humanity checks like worldcoin, cloudflare turnstile and others.
Well known accounts are disproportionately affected. Spammers try to get well known accounts to follow them so they can seem more credible. ZachXBT’s account probably sees 10x - 100x more spam than the average Farcaster post.
Spammers are highly incentivized. If you can create an account that gets well known accounts to follow you, a token that targets farcaster users might give you $100. The reward for passing spam checks is like 10x to 100x higher than Twitter, and the bad actors are much more motivated.
Warpcast uses ML to fight spam. We’ve developed an in-house spam model which tries to rank all counts into three tiers based on their probability of being a spammer. You can see this public dataset here: https://github.com/warpcast/labels.
Warpcast takes different actions based on how confident it is in a label. If we are very certain that someone is a spammer we hide their engagement. If its ambiguous we don’t hide them, but prevent them from generating notifications that may annoy other users.
ZachXBT reported a problem with a large number of likes and recasts coming from bots (cast).
@frostnymphs is one of the spam accounts that ZachXBT flagged as being problematic. let’s look at the history of this account.
@frostnymphs account was created on Feb 2nd at ~2pm.