The place mentioned is 99% fake posts now. Most posts there are karma farming.
I am sad for those researchers. What they got is mostly a map of reverse projections. It's a social game of chicken with virtues, not the real thing. And it's also useless.
moritzwarhier•8mo ago
The idea of AITAH reflecting "real life morality" is interesting, but indeed would only remain interesting for such a question when it also addresses LLM training and bot posts+comments, as well ad human fakes/ karma farming / spam.
Perceived morality is still interesting when it comes to the content, but ignoring this "dead internet" aspect is a glaring omission.
This subreddit and similar ones sometimes almost feel like ChatGPT "ground zero" to me, regarding the fine-tuning towards a chatbot for humans, and RLHF (vote system...)
By now it has become a feedback loop, I'd assume, and Reddit collaborates with AI companies, so maybe they even can separate extremely easily between obvious bots or manipulation and human posts. Also, they can already do that 90% on their own, given the platform design. E.g. detecting an essay pasted into an editor, IP ranges, etc pp.
There has been content and subs popping up in other languages that feels similar, for quite a while.
From hyperlocal (city) subs to topic subreddits, manipulation and LLM content is everywhere.
alganet•8mo ago
I am sad for those researchers. What they got is mostly a map of reverse projections. It's a social game of chicken with virtues, not the real thing. And it's also useless.
moritzwarhier•8mo ago
Perceived morality is still interesting when it comes to the content, but ignoring this "dead internet" aspect is a glaring omission.
This subreddit and similar ones sometimes almost feel like ChatGPT "ground zero" to me, regarding the fine-tuning towards a chatbot for humans, and RLHF (vote system...)
By now it has become a feedback loop, I'd assume, and Reddit collaborates with AI companies, so maybe they even can separate extremely easily between obvious bots or manipulation and human posts. Also, they can already do that 90% on their own, given the platform design. E.g. detecting an essay pasted into an editor, IP ranges, etc pp.
There has been content and subs popping up in other languages that feels similar, for quite a while.
From hyperlocal (city) subs to topic subreddits, manipulation and LLM content is everywhere.