Comment on Feature requests for a lemmy moderation bot
Jumuta@sh.itjust.works 2 months ago
I think the strike system would make sense and limiting by account age might work as well, but I don’t think the other simple method of moderation are going to be that useful, as word blacklists always have false positives and auto community lockdown could be exploited by someone to sabotage a community. (like with a denial of service attack)
Also, this is just my opinion but I really hate it when software speaks to me like they’re real people, I understand that it is ultimately written by a person but when the same phrase is repeated to every person regardless of the context it just annoys me. Similiarly I don’t like the automod messages/comments on reddit because it just feels like low effort content in comparison to what real people post on there.
I think this would make more sense framed as a plugin rather than a bot, because a bot post is placed equally to a human post while being inherently lower importance.
If I were to suggest features it would be something like a large language model based content filter, but I understand the computational and cost limitations would make that challenging.
AsudoxDev@programming.dev 2 months ago
LLMs are too expensive to run and slow as well. As for the lockdown system, DoS as in sending API calls to send messages? Yeah well that’s considered spam and will indeed lock the community down. But it won’t happen instantly and I still need to plan how it would detect spam and whether it will attempt to automatically ban the spammers or not. The bot is still in the preparation phase, I haven’t done any work yet.
As discussed in some other comment, the bot will not delete posts that are triggered by regex or word blacklist but report them. The bot indeed won’t understand context so it will do more harm than good. I am also thinking of transliterating content to catch people using symbols that look like certain characters like € for E and $ for S.