The advent of LLMs really opens the door to shunting off these “community members “ who’d rather contribute in misanthropic ways for the lulz than either leave or not contribute at all. They can take part in an interactive echo chamber that gives just as well as they can. You don’t even need a powerful model so the overall costs to the community are probably lower than the alternatives of trying to coexist with community-arsonists.
I spent years trying to find ways to bring people productively “into the fold” but eventually realized that it is futile in some cases because there’s zero value to the individual or the community to find a middle ground. They want to see things burn, and the community simply wants them out.
You're talking about using a machine to detect social undesirables then quarantining them in the matrix.
I’m specifically referring to people who have seemingly made it their sole purpose to create as much indiscriminate damage as possible.
You can ban them, block routes for them to attempt to Sybil themselves back to having accounts, etc. but even with great moderation tools and systems, it’s extremely difficult to set up a strong enough set of controls which don’t adversely impact everyone else who you want to have participate in the community.
Yes, a shadow environment is dystopian. It’s not my nature to want to even consider using one.
But we’re talking about privately run communities which also deserve to exist to serve their purposes.
So given the choice between anarchy which drives away people who contribute to make the community what it is and a shadow option for those actively working against its interests, I’ll consider the community first.
You may have misinterpreted my comment. I’m not suggesting you use LLMs as moderators. I’m talking about using LLMs as participant “members” of this shadow board to interact with someone whose account was flagged by a human moderator.
---
Can you please edit out swipes like "Do you hear yourself?". This is in the site guidelines: https://news.ycombinator.com/newsguidelines.html.
If you wouldn't mind reviewing the other guidelines too, we'd appreciate it. Note this one, for example:
"Please respond to the strongest plausible interpretation of what someone says, not a weaker one that's easier to criticize. Assume good faith."
Further it may still make sense to use human reports to gate some automation even if it slows response.
