I wonder what makes a platform like HN work, but not the others.
In almost every other platform moderators are just sad, angry little entitled narcissists who love exerting control over others. This has been proven time and again across multiple platforms:
Wikipedia
Quora
Stackoverflow (surprise, surprise!)
..
And basically anything else that depends on those so called moderators for fairness and equality. It would be interesting to experiment using an LLM with explicit set of hard guidelines (like outlined in the Reddit's code of conduct) and see how it behaves. Sure, LLM's are biased due to their training sources, but I'm curious to see if they will be as biased as human moderators. We need the HN formula for the rest of the platforms (I know HN doesn't use AI) with or without AI.
I've asked myself this many times. It warrants a study.
I have managed large sites where I had to recruit mods. I would recruit the most popular and lovely users to be mods, and universally I would be forced to ban them within about 6 months. The power would go to their heads and every one of them would turn into a fascist dictator just banning anyone who spoke out of turn and deleting any content they didn't like.
Dang. Dang makes this work.
What an insanely hard job, done with far more grace and far fewer mistakes than I could possibly pull off.
Thank you for this corner of the internet, dang (and a couple others).
Edit: mobile typos