Oddbean new post about | logout
▲ ▼
pippellia | 7 months ago (raw) | root | parent | reply | flag 
 I do have the code that do the thing.

https://m.primal.net/HxnG.png 

Also, I think Coracle WoT model is very naive. Why pow(2, log(n)))

2^ln(n) = 2^( log2(n) / log2(e) ) = n ^ (1/log2(e))

Wtf is that? It's ugly! it's arbitrary!
▲ ▼
mleku | 7 months ago (raw) | root | parent | reply | flag 
 i haven't actually thought about the math of it, and i kinda don't understand normal math notation i'm used to thinking about functions in code form

as i understand it, web of trust is about some kind of hamming distance type thing between nodes in a graph, and you sum all of the interlinking paths by an exponentially decaying weight - ie, the value for close nodes falls off exponentially after several

so i guess, thinking that through, that he is using an inverse square weighting, against the sum of all the paths between two users based on their follows

this is only one part of it too, i think that you can also add weight by the number and frequency of replies from one user to another to evaluate more than just the simple proximity, to actually how the path is evaluated by each party in the graph
▲ ▼
pippellia | 7 months ago (raw) | root | parent | reply | flag 
 There are countless ways to do WoT, some are cooler than others.

But what Coracle is doing is just strange.
▲ ▼
hodlbod | 7 months ago (raw) | root | parent | reply | flag +1 
 Yeah, I'm not a math guy. The goal was to weight mutes against follows in a reasonable way. Someone suggested I use tan instead. The nice thing is no one has to use the same formula.
▲ ▼
pippellia | 7 months ago (raw) | root | parent | reply | flag 
 yeah, maybe I was a bit harsh, but the thing is that as you can see from he math above it's just strange.


if N is the number of people you follow, who follow person X

and n is the number of people you follow, who muted person X

Then why not just 
WoT = N - n

now you are using 

WoT = N - n ^ (1/log2(e))

which is just strange. There is no good reason behind it.
▲ ▼
hodlbod | 7 months ago (raw) | root | parent | reply | flag 
 No, it's because I want to underweight small numbers of mutes, and overweight large numbers.
▲ ▼
pippellia | 7 months ago (raw) | root | parent | reply | flag +3 
 If you are interested, we can have a call where we can talk about interesting approaches for WoT.

I am a mathematician super interested in WoT, and I've read almost everything I could find about WoT in the last 3 months 😅

@npub1jlrs53pkdfjnts29kveljul2sm0actt6n8dxrrzqcersttvcuv3qdjynqn
▲ ▼
hodlbod | 7 months ago (raw) | root | parent | reply | flag +2 
 Let's do it, want to come on nostr:nprofile1qy2hwumn8ghj7un9d3shjtnwdaehgu3wvfnj7qghwaehxw309ashgmrpwvhxummnw3ezumrpdejz7qguwaehxw309a3ksunfwd68q6tvdshxummnw3erztnrdakj7qg4waehxw309a3k7unpvdkx2tnnda3kjctv9uq36amnwvaz7tmwdaehgu3wvf5hgcm0d9hx2u3wwdhkx6tpdshszxthwden5te0dphkgmrzdajzumn0wd68yvfwvdhk6tcpzamhxue69uhhyetvv9ujuurjd9kkzmpwdejhgtcpz4mhxue69uhhyetvv9ujuerpd46hxtnfduhszymhwden5te0wp6hyurvv4cxzeewv4ej7qpqmlcas7pe55hrnlaxd7trz0u3kzrnf49vekwwe3ca0r7za2n3jcaqta6u9k ?
▲ ▼
pippellia | 7 months ago (raw) | root | parent | reply | flag +1 
 That would be cool as fuck! Thank you

sending you a DM
▲ ▼
Aida | 6 months ago (raw) | root | parent | reply | flag 
 I'm not a math guy either, so I can't really imagine WoT working just from within. Maybe as a recommendation tool for interesting npubs with similarities to you, but no idea how it will work for newly onboarded users. What would they see? How would they be seen? Would it make sense? 

I recently proposed NIP-85 https://github.com/nostr-protocol/nips/pull/1210 to create chains/networks of trust between NIP-05 providers. It is not based on calculated trust, but real consideration of compatibility.