cross-posted from: https://lemmings.world/post/21993947
Since I suggested that I’m willing to hook my computer to an LLM model and to a mastodon account, I’ve gotten vocal anti AI sentiments. Im wondering if fediverse has made a plug in to find bots larping as people, as of now I haven’t made the bot and I won’t disclose when I do make the bot.
What I am failing to understand is: why?
Is this just for some petty motivation, like “proving” that people can not easily detect the difference between text from an LLM vs text an actual person? If that is the case, can’t you just spare all this work and look at all the extensive studies that measure exactly this?
Or perhaps it is something more practical, and you’ve already built something that you think is useful and it would require lots of LLM bots to work?
Or is it that you fancy yourself too smart for the rest of us, and you will feel superior by having something that can show us for fools for thinking we can discern LLMs from “organic” content?
People do things for fun sometimes. You could ask this about almost anything that people do that isn’t directly and immediately related to survival. Why do people play basketball? It’s just pointlessly bouncing a ball around in a room, following arbitrary rules that only serve to make the apparent goal of getting it through the hoop harder.
This is not the same as playing basketball. Unleashing AI bots “just for the fun of it” ends up effectively poisoning the well.
No, it’s not the same. I was using basketball as an analogy. Someone who doesn’t enjoy basketball wouldn’t “get it”, just as you’re not “getting” the fun that can come from building and playing around with AI bots. Different people find different things to be fun.
I completely understood your analogy, and I certainly understand the fun in tinkering with technology. What you might be missing is that it seems that OP is planning to write a bunch of bots in here and then test how well people can detect them, and that affects other people.
Right, and this is presumably something he finds fun. You were asking why, I was explaining why.
Why would be easy to understand. To implement counter AI measures, best way to counter AI is to implement it.
Btw I like the Lemmy hostility
You are jumping into this conclusion with no real indication that it’s actually true. The best we get for any type of arms race is a forced stalemate due to Mutually Assured Destruction. With AI/“Counter” AI, you are bringing a medicine that is worse than the disease.
Feel free to go ahead, though. The more polluted you make this environment, the more people will realize that it is not sustainable unless we started charging from everyone and/or adopt a very strict Web of Trust.
Well the only way to know I’m write or wrong with certainty is to test myself. It’s like we both don’t know if I’m right. I’m going to figure it out because I’ll investigate and I’ll have fun investigating.
The alternative is we don’t do anything about the LLM bots on the fediverse and they just integrate in. Also there’s only one way to see if your MAD is correct
You don’t do tests in an actual production environment. It is unethical and irresponsible.
Feel free to do your experiments on your servers, with people who are aware that they are being subject to some type of experiment. Anything else and I will make sure to get as many admins as possible to ban you and your bots from the federation.
How’s it unethical?
You want to write software that subverts the expectations of users (who are coming here with the expectation they will be chatting with other people) and abusing resources provided by others who did not ask you to help you with any sort of LLM detection.
That doesn’t answer my question and it’s not coherent. Like I’m apparently “abusing resources” when I use a bot but not when I use a bot to make a leaderboard that tracks fediverse streamers stats or if I make the content with fleshy brain, just my resources.
There is a big difference if your bot provides functionality that is good for the community and a bot that does only things that interests you.
People are asking you not to do this. So, if you want to do it, do it on your resources. I’m saying that as someone who set up almost 20 different instances (alien.top + the topic specific instances) just to have a place to run the mirroring bots.