I moderated a few chat rooms and dreamed of all this complex tooling I could build to automate and help me fix issues right away. Of course, the tiny chat rooms didn't need any of that. The hard part is actually moderating fairly and trying to figure out where the boundaries are. I couldn't do it. Makes sense why pg said "don't start a forum".
I know they get an alert whenever a comment is flagged. I assume they get an alert whenever certain "problematic" accounts post. Maybe they get an alert whenever the flamewar detector goes off? This forum is written in a Lisp so they can probably hotpatch whatever code they want and add whatever filters they want.
But dang has said numerous times that he doesn't read everything, that would probably not even be feasible.
The question is, YC being what it is, are they using LLMs to automate moderation or do sentiment analysis?
We're working on the latter but it's not yet in production. We're only going to deploy it once it's clear that it significantly improves on what mods + the existing software are already doing. The hard part, of course, is how to evaluate that.
Not even close.
https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...
I moderated a few chat rooms and dreamed of all this complex tooling I could build to automate and help me fix issues right away. Of course, the tiny chat rooms didn't need any of that. The hard part is actually moderating fairly and trying to figure out where the boundaries are. I couldn't do it. Makes sense why pg said "don't start a forum".
I know they get an alert whenever a comment is flagged. I assume they get an alert whenever certain "problematic" accounts post. Maybe they get an alert whenever the flamewar detector goes off? This forum is written in a Lisp so they can probably hotpatch whatever code they want and add whatever filters they want.
But dang has said numerous times that he doesn't read everything, that would probably not even be feasible.
The question is, YC being what it is, are they using LLMs to automate moderation or do sentiment analysis?
We're working on the latter but it's not yet in production. We're only going to deploy it once it's clear that it significantly improves on what mods + the existing software are already doing. The hard part, of course, is how to evaluate that.
A person would lose their mind pretty quickly this way.
Great potential job for an LLM to potentially spot abuse. The existing mod tools/system seem to work quite well.. I like HN how it is!