People have very different thresholds for type i and type ii errors.

For example, I spoke with many about moderation this week, some would much rather let 50 spammy posts thru than block 1 real one. Some closer to the opposite.

Challenging to design for. It's the intersection of ethics and practicality.

building /automod (channel bots), /glass (no code frames) • uncurated.xyz
• prev @manifoldxyz, twilio

learning as i go but i'm sure machine learning folks are very used to bumping up against this stuff.

interesting to consider everyday situations too, for ex: justice system would rather let guilty ppl free than convict someone innocent (min. type 1)

more generally still:

learning as i go but i'm sure machine learning folks are very used to bumping up against this stuff.

interesting to consider everyday situations too, for ex: justice system would rather let guilty ppl free than convict someone innocent (min. type 1)

more generally still: https://en.wikipedia.org/wiki/Decision_theory

i was hoping this was a thread

...or an article

... or a podcast

you probably know a ton about this subject now

building /outpaint | experimenting onchain /artivilla /pirates | ex-coinbase ex-bisontrails | OSS noob

Yep, you're in the minimize "type ii" errors camp. You'd rather see a bit of spam than risk missing a legit cast.

There's an extra dimension at product scale too: you'll need some policy and to choose one its often a trade off of simplicity (majority, one path) and complexity (configureability / branch factor).

I would rather block users and then allow list them through a form. And if the users overtime leans spam (which I don’t think is the case) they are re blocked or hidden from feeds. Quality of life is really imp and growth through word of mouth won’t happen on spam filled networks.

it’s a vast gray area, not black and white

I assume it’s what moderators are optimising for, i.e inclusive vs exclusive, growth vs retention.