Ask HN: Forbid Reddit HN Submissions?

[nateb2022]

cc @dang

Reddit has been filling up with AI generated content. Either bots engaging with naïf audiences who don't realize they're bots, or supposed humans who defend their giant copy/pastes as somehow justifiable as a spellchecked/proofread version.

I don't believe AI generated slop posts deserve consideration, either on HN or Reddit. If people rather not take the time to write something as long as these AI outputs, they're free to write a prompt-length post instead of forcing us to read the AI equivalent of a zipbomb. We can't control what people post on Reddit but I think we should have higher standards here and that Reddit has gone to the point that only very very selective submissions from there should be accepted.

On a tangent, does anyone know of a browser extension I can use to easily hide or at least highlight in red likely AI generated text?

[sometimelurker]

I've been thinking about the bot problem: it makes it much easier to develop better bots if theres a clear signal if they're seen as humans or not. So its really not worth it to visibly say "xyz is a bot", as that would make it easier for the people making them to iterate.

About the browser extension, it might work a lot better if you made it, otherwise it would just become a training target for people making bots. I think a (-n overkill) safe way to check if something is AI is to just check the account creation date, and only read further if it's before 2022. That's what I do on political posts here anyways.

[hootz]

We can't control what is posted there, but we can control what is submitted here, no need to ban Reddit submissions, just check the content.

[Bender]

dang wont see this message unless it gets a lot of activity. best to email him, contact at the bottom of the page if you have suggestions how to better detect and auto-[dead] the junk.

We are flagging the AI slop but there are times of the day that not enough of us do it. The botters know people have to {sleep, do IRL things}. Best to just flag it and move on for now but dont go out of your way to do it. The sites that cant manage it will eventually just fade away and become ghost towns. Ghost towns always have a handful of people that stick around.

FWIW I know how to spot the bots on my hobby sites before they post their first byte but it can block legit search engines if people care about such things (I don't). (Blocking HTTP/1.1 connections, the long list of user-agents because most botters are too lazy to change it ... oh yes they are and most will not spoof, blocking TCP SYN missing things in the header, blocking connections missing some variables such as $http_sec_fetch_mode and others, any connection coming from a data-center and this is what nukes some search engines)... Just dropping HTTP/1.1 connections will drop about three fourths of them but that also nukes the GoogleBot. Would be nifty if Google could fix that. BingBot is HTTP/2.0.

[krapp]

Practically everything everywhere is partially or entirely AI generated now, certainly most things likely to be posted here, and comments here are increasingly the same. Focusing on any specific platform is pointless.

There is no effective way to escape it or filter it, although the mods try and the guidelines are adamant that such content isn't welcome here. This is a forum hosted by a Silicon Valley startup incubator formerly run by Sam Altman after all, so it's doomed to be the canary in the coal mine for AI content accelerationism. One might as well be King Canute commanding the ocean of piss not to bring the tide.

> On a tangent, does anyone know of a browser extension I can use to easily hide or at least highlight in red likely AI generated text?

There is no way to easily and effectively detect AI generated text as far as I know. All of the "tells" people think work are basically witchcraft. There are tons of extensions and plugins but they'll tend towards false positives. Dang is working on something on the backend to detect generated comments, and I'd be curious to know how well it works but for obvious reasons I doubt he wants to give specific details.

[chistev]

Let users downvote and upvote as they see fit.