Comment on Is the bot/troll situation getting worse?
blahsay@lemmy.world 11 months agoIf I was designing an anti troll/bot system I’d implement a few things. Let’s call any bad actor on here a bot/troll or broll for ease.
You can check if text is ai/llm generated and an automated api check before posting and immediate ban if found
If this LLM-detection function ever results in false positives, this system will be banning innocent people.
Also there are many, many cases where a person openly displays results from an LLM, without it being in any way antisocial.
The odds of someone coming up with the same sentence as an llm within common sense bounds of time far exceed winning the lottery or getting struck by lightning.
Your second point is straight up nonsense. This platform is for humans to interact. The use of bots is inherently deceptive.
Fascinating to have someone argue for them. I think the backend logs will be pretty illuminating.
I don’t know what a person “coming up with the same sentence as an llm” would have anything to do with this unless the LLM detection is based on direct string comparison.
The use of bots is inherently deceptive
Nope. I can say:
Here’s what GPT-4 generated when I gave it that prompt: “[some LLM output that would get them banned by the machine we’re proposing to build]”
That is not deceptive. But it would be detected by this system and result in them being banned. Because you guys are gung-ho to build a powerful head-cracking machine and didn’t think of an obvious edge case.
You’re wrong and don’t have the technical knowledge to understand why and I can’t be asked explaining it.
Relax, it won’t affect that case.
haui_lemmy@lemmy.giftedmc.com 11 months ago
Very good ideas! Any idea if something like this already exists? If not, shall we work on something? I have some experience in python if that helps.
blahsay@lemmy.world 11 months ago
Thanks but not sure what’s currently implemented or even what the code base is written in 😅.
I might have a poke around and see if there’s any low hanging fruit.
Call me crazy but with a 5b ipo about to start I’d be shocked if reddit wasn’t paying some troll farms to brigade the fediverse and it’d be a shame if spez wins
haui_lemmy@lemmy.giftedmc.com 11 months ago
Thats an interesting idea! Thank you very much for mentioning it!
We can absolutely write a bot in python and could try to use it like that. I already made a discord bot so this shouldn’t be brutally hard.
blahsay@lemmy.world 11 months ago
Awesome 👍 I’m more c#/Java, angular if there’s anything I can contribute.
rimu@piefed.social 11 months ago
PieFed is an open source lemmy alternative (written in Python) that makes good use of karma/reputation, as shown in this video:
https://mastodon.nzoss.nz/system/media_attachments/files/111/648/646/494/228/522/original/02cb1b5182a1f9b6.mp4
Try the demo site at https://piefed.social and check out https://join.piefed.social. Also see https://piefed.social/c/piefed_meta for recent feature announcements.
Blaze@discuss.online 11 months ago
Thank you very much for sharing, I’ll keep an eye on it.
haui_lemmy@lemmy.giftedmc.com 11 months ago
I‘m not searching for another thing to start but a way to make the current thing work. But thanks.
Blaze@discuss.online 11 months ago
To be fair, Piefed uses Lemmy communities and comments, it’s almost just another interface.
The reputation is indeed interesting, example in this thread with warnings “low reputation, beware!”: piefed.social/post/27070#post_replies