🤔
Hmmmmm.
Discussion
Hmm indeed. What you saying semi
WoT is broken has been always will be
we have spambots like nostr:npub10u8xfdfw7447c26c34rql333yh6k0kevq9x3an8gqmvdtdpqnchqpmz5qy that have high WoT scores, and AI slop makes people easy to manipulate
report-based systems can and are being exploited because people can’t agree to disagree
So we give up? Cause you tell us to?
no, I am providing my opinion here.
but it has been repeatedly shown time and time again that WoT does not work, and if it was such a good system, it would have already seen major research done into it for the decades that the concept has existed
You talking about the new amethyst thing?
not specifically. WoT is flawed and I have been calling it out for a long while.
Flawed concept or execution? Given it's configurable and optional, and to me the idea seems sound, I don't see the problem many seem to be expressing today.
Is the issue that some large client rolls out a default setting that many users are too lazy to configure or disable?
concept
Oh, interesting. Think I've seen you mention the gaming aspect is too easy, but that seems to be more a flaw with easily-fooled users. I'd expect one could configure to only score based on a selection of the people they follow, who one believes do a good job of filtering their follows. That's basically what I attempt to do today when I stumble across a new npub; look at shared follows, scan for the ones I know apply a rather strict criteria. Works aight for me, mixed in with a little timeline review. I do really hate the scammy types of accounts out there, but that's only gonna get worse without some type of WoT I'd imagine.
Guessing opinionated relays is your fix? Less reliance on follow behavior and more on whatever algo you use or intend to deploy on Nostrland.
Just updated Amethyst from zapstore. What feature are you talking about?
That npub is 0 for me 🤓 Blocked.
the problem is anyone can turn up such a AIslop account, make it like posts/follow en masse, get followers and then reuse it for spam. rinse and repeat
GrapeRank is t just based off of likes and follows though. And it's configurable for each user. It's spam resistant because of its configurable nature.
an algorithm which uses inherently flawed primitives is flawed no matter how much :
- mutes/reports: inherently personal, and can be used by attackers
- follows: content interest != reputabulity
- likes/engagement: even worse than follows, someone could spam and you only see 1 good post you like
the current algorithms are too shallow and assume a relatively static environment where everything has already been bootstrapped
WoT score on Coracle -3 😑, that's the simplest metric
Do you need help, guys? 🖖