Reddit is using AI to determine users beliefs, values, stances and more based on their activity (posts and comments) summarizing it to Subreddit Mods.
cross-posted from: https://programming.dev/post/36063172
A black box you can't look inside making determinations that could affect your access to the entire site. How dystopian. It's also pretty clearly only useful for pruning users en masse and not have to bother with the responsibility a moderator should be subject to.
Yeah, so this was always going to happen eventually, right? It's yet another reason you're not anonymous online unless you actually are.
Next stop, AI-powered doxxing.
And it's not like this is new, Redditmethis (and similar things) has been a thing for, oof, over a decade? Well before all of this AI stuff.
https://redditmetis.com/user/GallowBoob
(If it doesn't load, enhanced tracking protection breaks it because it needs to cross-connect to reddit)
Yes, doxxing has always been possible. When I think about how you could dox this account, though, it's not ways that would show up on a simple tool like that. An LLM might be able to piece together a chain of thought for it, though.
At this point I think the cat is out of the bag. The internet isn't private anymore. At the same time I question what happens if we start to self censor. On one hand I think it keeps a person safe, on the other hand I think it could just start to let authoritarian or fascists take a stronger hold when they're the loudest voices.
I mean, you can still be anonymous, but it's mildly technical and a lot of extra work.
Most people won't bother, though, you're right. And that's not great, although not different from before the internet, but then again a system like this is a lot cheaper than a human Stasi agent which is scary.
And i continue not regretting deleting everything i ever posted on Reddit for 17 years when i left them during the APIcalypse fiasco (yes, i imagine it's all archived somewhere anyway, but at least it's not there for them to use at this time)
The reddit API can only go back 1000 comments (or pages of comments, 25 per page, I forgot which). I went back to about 2019 fairly well, but none of the scrubbing scripts really accomplished anything beyond that. How were you able to delete all your comments and posts?
It's been a couple of years so don't recall the details or the name of the thing, but it was something that went through every comment you ever posted and overwrote it, even past the 1000 messages limit, but got crippled after the API stuff actually got implemented, no idea if it still exists or if it's usable
Even if you delete it, they just restore it. You do not own any of the content that you posted. You signed right away when you agreed to become a redditor.
I also noticed that the youtube app recently started giving "help" about what comment to write. That small area where you can comment on a video, it showed something like "what's the best band you heard? 🤔". I'm not sure if that is a new thing creators can add, but with the think smiley it feels very much like AI". I don't need AI to tell me what kind of comment to give. It feels like AI is trying to dilute my thoughts.
I recently watched a video where they tested if the replies were generic/context-generic or fed by their own comments.
They replied to comments on their video with a fake ad. And sure enough, after a while, the AI suggested such replies.
Thanks I enjoyed that video
Absolutely does.
I dont actually know. I am just so certain that it does
I was going to say that this might actually be a good idea, but then I remembered how often AI summaries get key details completely the wrong way around.
I still think it's a good idea in concept: modding is hard, or so I hear, and this could be a very valuable tool if it worked as it's supposed to. I doubt it would.
There are some details that I'd have done differently too, if I was somehow forced to implement something like this. For example, for each point in the summary (e.g. "talks about US politics") there should be a list of comments/posts by the user to exemplify what they say and how they write. It's useless if the summary doesn't show you what it's based on.
Lol, you can tell which commenters have never moderated anything in this thread, IMO. If it weren't for the high likelihood that these summaries will be wrong an appreciable percentage of the time, this would be a huge help for anyone moderating medium traffic subs. Those types of subs, especially if they have relatively hands-on moderation to keep them from being complete cesspools, often involve seeing a comments or post that is borderline, and feeling like you need to go look through the poster's history to figure out if they're a bot or a troll. Something like this that actually worked, especially if it linked back to a sampling of the posts/comments that it is referencing, would be a big help in that. Also something like this that summarized a user's moderation history would be pretty useful.
Right, here’s my rough process:
Along the line, discuss with other mods in real time, off platform because Reddit is still not properly accessible.
Anywhere along the way I may feel confident to make a call and skip the rest.
And I only deal with a 30k sub…
That said, Reddit’s already filtering things for likely spam that just aren’t. I don’t have huge confidence in them and I don’t have any confidence in LLMs for this.
Nah. It's just creepy.
So glad I deleted my account
i couldn't delete mine and i wonder what it says about me.
What the fuuuuuck 😳
I can see how that could be a useful mod tool, but still (see line 1).
So happy I left that hellhole 😮💨
Edit: Although I guess this means I should have invested the time in deleting all my posts; probably too late now 🙄
And damn, now I really want to know what their AI thinks of me! 😅
Can we talk about the emojis for a minute..?
This isn't reddit, there's nothing wrong with emojis.
I'm sorry; it's a coping mechanism for not being able to adequately read mood/tone from written text. I'm trying to quit, or at least dial it down, but it's a struggle 😞
With my current progress on that front, we might as well be talking about grammar and punctuation — right ...? 😶
Yeah, those should have been emoticons! (╬≖_≖)
That sounds like a tedious conversation, even for a minute.
Overall, I can see liking this. But mostly I think the summaries should be public.
In general the problem with moderators is they can be fairly partisan. I don't know if it's still the case after the whole API ... thing, but certain groups of moderators had access to bots that did what is essentially equivalent to the sort of thing. What tended to happen is the good mods would become overwhelmed and bring on a "power mod" and the powermod has secret axes to grind and political agendas that they bring with them into the sub.
Another problem I generally had with reddit towards the end of my serious engagement there is that a lot of things reported to admin started being evaluated by non-American English speakers who don't have the cultural context to understand sharp turns of phrase and plays on idioms. Americans would understand the words mean the opposite of what they literally say, but you can't expect ESL overseas contractors to understand these nuances. So I would be concerned that AI is similar... except for the fact that it's not really a change from the status quo.
Would be nice if we also had AI summaries of moderator behavior and if these were visible to everyone. I wouldn't be surprised if admin have (or soon will have) access to AI summaries of moderator activity and behavior. So that might be another shoe to drop. I can see it can be good if it improves modding for the good mods who just want to build communities. Basically it might reduce the need for the powermod protection racket.
Speaking as a reddit mod for a massive subreddit, it honestly hasn't been a useful tool at all. It's only really useful for identifying users who engage with problematic subs but there are old tools which can do that without using AI.
I kinda want to see what nonsense it says about me.
When the summary is presented there, like that, of course, it seems helpful.
I'm certain the majority of people will respond with "yes", without checking the user's post history to verify the summary. The responses will be biased, making the whole thing worse, and giving opportunity to disingenuous marketing and justification.
We can only hope they interpret and use it for that: indicating usefulness, not correctness. Who knows what kind of metrics they use for correctness, if at all, and what type of faults they deem acceptable.