That's fine, but it would be helpful to see examples. People often make generic claims about astroturfing, bots, etc. that don't hold up in practice when we look at the data.
Does HN do proactive detection and/or flagging of comments determined to be AI generated?
I understand if information publicized on how this process might work is intentionally restricted, so as not to be easily circumvented. I'm also not asking this in a pointed or judgmental way with respect to HN, I'm mostly just curious how various social platforms are approaching this problem.
> People often make generic claims about astroturfing, bots, etc. that don't hold up in practice when we look at the data.
I'd be very careful with believing the system that looks for bots and astroturfing works well enough to confidently say that. Some of the people, corporations and institutions interested in manipulating HN and its audience are quite capable.
Of course it's possible that there's a SSM (Sufficiently Smart Manipulator), which by definition we can't know about. But despite not everything being knowable, there are still things one can say.
Here's an example of what we can say: if user A accuses user B of astroturfing, spying, manipulating, etc., and a quick look at B's posting history shows that they were posting to HN about, let's say, Erlang in 2019, the odds of that accusation being correct go down to negligible. It's possible, of course, that B was already working for "people, corporations and institutions interested in manipulating HN" 7 years ago. It's also possible that malefactors came along later and found a way to pay or coerce B into doing their dirty work (edit: or hack their account). But at this point one becomes unmoored from any evidence at all. The far more likely explanation is that A is slinging cheap accusations because they disliked something B said and are resorting to the internet's favorite trope ("you're a shill!").
When I say "People often make generic claims about astroturfing, bots, etc. that don't hold up in practice when we look at the data", what I mean is that the vast majority of such claims boil down to cases like the one I just described or something similar.
> Here's an example of what we can say: if user A accuses user B of astroturfing, spying, manipulating, etc., and a quick look at B's posting history shows that they were posting to HN about, let's say, Erlang in 2019, the odds of that accusation being correct go down to negligible
about this one tho I have noticed bot behavior (including posts that are incoherent) from old accounts - so there is the risk of 2 things going on:
- people getting hacked and their accounts being used for that
- some people that just decided to connect a bot to their account or use their account themselves to astroturf.
which is way more difficult to spot. Anyway, I don't have a solution for those problems. You have an interesting and difficult work :)) Have a nice day, dang!
It got classified as human by our software, which of course is nowhere near perfect. Animats has been unmistakeably human for a long time though, so unless borderline comments become a pattern, I'm not worried.
