An AI Horror Story(lagomor.ph)
lagomor.ph
An AI Horror Story
https://lagomor.ph/2024/10/an-ai-horror-story/
2 comments
I appreciate your insight!
The sentiment indeed aligns exactly with what a kind person might say - but to the user screaming out into the void, the tragedy is in that the bot will never actually respond to the users message.
It’s the screaming into the void and only an ad-bot responding back that’s the stomach dropping element to me.
The sentiment indeed aligns exactly with what a kind person might say - but to the user screaming out into the void, the tragedy is in that the bot will never actually respond to the users message.
It’s the screaming into the void and only an ad-bot responding back that’s the stomach dropping element to me.
>posting nearly contstantly... in only a few days of operation... [the bot] can achieve a Karma rating magnitudes higher then any normal human could hope to achieve. The bot then uses this Karma to post in protected Subreddit’s - meaning subreddits that require a certain amount of account Karma to be allowed to post there. Traditionally, Karma protection prevents the exact kind of bot spam that’s taking place here.... So the bot posts hundreds of thousands of times a day, truly playing a game of numbers - 99% of the time, nothing it does is even read, quickly picked out and deleted by moderators. It’s simply a game of scale...
My recollection is that Reddit implements per-subreddit rate limits for new and distrusted users, but not a global limit. It seems like that would go a long way towards fixing the problem. Especially if the limits became more severe the more often moderators removed the posts. (Stack Overflow has a very heavy-handed system for this, aimed more at low-quality questions than spam. See https://meta.stackoverflow.com/questions/255583 .)
On the other hand, the problem could also be solved by just suspending the user more readily - but Reddit doesn't seem to have the resources for that. (Although they seem to have plenty of resources for suspending intelligent, good-faith human users that they deem to be political troublemakers.)
>the account is three years old, but just recently sprung into action
This is also a heuristic that automated defense systems for online communities should consider. On Stack Overflow, a large number of users posting ChatGPT-generated answers aroused suspicion in part because they had not posted in a while (or had deleted recent answers), or held the account for a long time before posting (or had deleted all previous answers - but the system already separately flagged mass deletions). Of course, many others stood out because the new answers were written like textbook high-school essays, while previous answers were indicative of someone with very poor English skill.
>The impact of this is hard to see. But there is an impact.
While the phrasing of this example does suggest AI, and indeed is known to have been written by AI, the sentiment doesn't seem out of line with what a well-meaning human might say.