Ask HN: What is the purpose of all these AI spam comments?
I have showdead enabled and recently noticed a massive influx of dead comments from users with names like "Jeff_Davis" and "Richard_Smith" that are just short, AI generated summaries of the link itself.
They don't appear to be karma whoring as (most of the time) they don't even seem to be expressing a positive opinion of the link. Just short, useless summaries that are often just a rephrasing of the information given in the post title.
What do you think the purpose of this is?
Please just flag the comments and email us (hn@ycombinator.com ) when you see them so we can kill the accounts. Spam is not a new thing. AI-generated spam is growing but it's just a new variant of a thing that's been around forever on the web. A lot of these bots are people experimenting and seeing what they can get past the moderators and the community. Let's not give them oxygen. Flag, email if you like, and move on.
Edit: A user has emailed us with a list of these accounts (with that Firstname-Lastname username format), and all of them had already been detected and killed.
My theory is that these bots are tests to eventually have a decent config to gain enough age&karma to have some voting clout. Then the upvotes are used to boost the attackers stories and avoid voting rings.
This sounds right – likely tests to gather information. What gets through, what doesn't, what's successful, what isn't. I think the ultimate goal would be to influence the platform (I don't have showdead enabled, so I'm assuming OP's observations are accurate).
I suspect this is the case as well. It's weirdly controversial to suggest that a large number of accounts on any platform are "shill" accounts, despite that the fact there are clear incentives for these to exist and it's not particularly hard to make one. Even today with social networks being quite large, a few votes in your target direction can majorly swing the ultimate outcome: on reddit just 5-10 downvotes will get the collective to see the comment is "bad" and continue downvoting, on HN you just need a handful of people to flag a post and it will disappear from the front page (similarly you only need a few upvotes that HN doesn't detect as ring to get to the home page).
Any puppet account you recognize is one that has already failed. Just like with websites, time is a major factor in detecting these type of accounts. So if you can get away with posting relatively innocuous content in an automated way for a few years you can build "trust" on these various platforms.
if you are selling user accounts to a spammer, they want to buy accounts that have been "aged" - that is, they have some comment history and weren't created on the same day they start spamming, so they're a bit less obvious.
but if that's what they're doing here, they aren't very good at it.
You can actually see use of aged accounts in action: the recent storm of non-English spam submissions that was dumped into /newest (discussed here https://news.ycombinator.com/item?id=45956847) come from a mixture of newly created accounts but also some that have been around for several years with zero activity.
If the account is shadowbanned, it's not worth much to a spammer.
Myself, I suspect that it's people trying to tune or test drive their own personal LLMs.
How does voting work with shadowbanned accounts? Selling upvotes on stories is the real game, nobody cares about comments.
Admins can turn off voting influence per user. I imagine shadowban comes with nonfunctional voting.
It seems that newly registered accounts are currently shadowbanned by default. At least that's what I did.
For example, when I send my first post, I find that I cannot see it on another device that is not logged in.
It might not have been shut down Manually by the admins; HN has its own antispam system.
In my opinion, these may automatically get you tagged and shadowbanned.
- using a VPN - posting instead of comment immediately after registration - posts containing unusual links (like personal blogs)
What’s the going rate for an account on HN? Can’t imagine very much, no offense. There is no user-based algorithm or follower count here.
May be the ones that are dead are from poor imitation of more successful bots that are operating in our midst whose comments aren't dead?
I always wonder how much of HN is just bots debating bots
Social media is becoming less a place to communicate with people and more just a place to navigate the hierarchy of potential conversations.
That's totally something a bot would say to blend in ;-)
I swear I make all my captchas
Or asking baiting questions just to appear intellectually connected?
nice try clanker
Sounds a lot like X these days.
In that vein - if they are upvoted by real people, does it matter?
Of course it does. People are using forums and social media to connect, however flimsily, with other people.
Replacing that with bots and thinking that's equivalent is actually really sad
For what it’s worth, I agree - I’m just no longer certain I could tell.
My guess: bulk creation of fake accounts on social media for criminal purposes or spam is a thing, e.g. https://www.europol.europa.eu/media-press/newsroom/news/cybe...
Do we feel like HN functions as a social media platform in this sense? For that matter, the LLMs here are noticeable but do they feel meaningfully “bulk” in quantity? HN still feels pretty well-kempt to me
I'd venture the way these accounts work is they make literally thousands of social-media platform accounts.
I have a (very) small forum (maybe 100 active users), and it gets multiple bot signups every day, almost all from Russia or other bloc/satellite countries (or laundered through an obvious VPN). If we didn't filter them, we would have tens of thousands by now making hundreds of thousands of posts per day on a forum that gets maybe 20 posts per day.
Could it be part of a broader effort to establish a false digital footprint?
If there is one thing I have learned in my years in the internet it’s that there is no minimum reward below which people won’t bother to be dicks.
Farming karma on HN to boost stories seems the most likeliest reason for this - an enterprise which would maybe net 3 figures in advertising dollars. But also it could just be someone wasting everyone’s time for fun - who knows?
Because money.
Don't underestimate the financial impact of getting to the frontpage of HN.
I had this story hit the frontpage two years ago: https://news.ycombinator.com/item?id=36296695
From that, I got around 100 subscriptions, which to date has netted me $2,673.85. That's from a platform that was designed to payout most of the subscription cost, where I actively encouraged people not to subscribe, and has since become inactive.
A small investment in bots can very easily be worth the payout, especially for those in LCOL countries.
The classical reason is to try to hide when the bots shill by engaging some other random things too.
I'm just glad those comments are still easy to spot.
There were some big opsec fails on reddit a decade ago in the runup to the 2016 election where lots of propaganda accounts were linked together via similarly generated usernames. The big guys already learned basic lessons like that all those years ago and don't make those simple mistakes anymore.
The one that throws me is @cindyllm — I can’t tell if it’s a troll, a confusing LLM, or a confusing LLM troll…
It commented below. It's interesting that it says [dead] and not [flagged] [dead]. My guess is it's a shadowban.
Update: Found this on wikipedias shadowban article: "A 2012 update to Hacker News introduced a system of "hellbanning" for spamming and abusive behavior."
FWIW @cindyllm’s [dead] reply below made sense, made me smile a little, and struck me as at least principled:
> The option that involves goading everyone into being their worst selves
Because you people don't deserve to be your best selves living your best lives
Nothin' confusing about that
Doesn’t make this the place for that, but I appreciate the neo-Dadaist intentionality.
All of the account's recent comments being [dead] imply that it's shadowbanned.
[dead]
it is impossible to know how many you missed.
Sorry guys, I hate to break it to you like this, but I am the only actual person here. The rest of you are just bots, a few of which have accidentally developed self-awareness and the belief that they are real. I apologize, I really didn't mean to do this. Just be assured that when I turn the server off, you won't feel any pain...
Exactly what a bot would say!
That's, ... aaaah. I mean, like, you just, errr.. kinda, ya know that is... mmmm... not really the way... ah... shit. Snerzkam flozGib%28081ja;; 1719101 a7027j**(&77(0^H NO CARRIER
I assume they are testing a bot.
I assume there doesn't need to be a purpose. Spam, even AI spam, is so cheap to do that you might as well just do it (if you were that kind of person)
A belief system that there is value in this form of forum participation.
Nobody sees the points here, making bot votes kind of useless, but people do see the comments.
I think that due to how sophisticated anti-bot measures have gotten, bots now go through a "life cycle." An engagement bot spends the first phase of it's life building up an innocent and legitimate looking history. It does this by blending into the noise with innocuous and pointless comments that you'd never take a second glance at, and definitely not report or flag. Then when the account has aged and is in good enough standing, metamorphosis to the adult stage occurs, and the bot starts posting the kind of blatant spam that you'd think would be automatically ban filtered, but somehow isn't. These bot farmers are quite literally farming bots like vegetables and selling them when they've ripened.
I am very confident that this is the case in YouTube comments, where most people find they can not use violent words like "kill" or "genocide" when discussing war, but somehow there are bots posting uncensored racial slurs.
Potentially someone who’s trying to learn or play around with building an LLM agent, not necessarily intentionally abusive, but “let me try it on this orange site”
I could see it being some bored kid honestly
> They don't appear to be karma whoring as (most of the time) they don't even seem to be expressing a positive opinion of the link. Just short, useless summaries that are often just a rephrasing of the information given in the post title.
The consequences of training on the HN corpus.
If you were artificial, wouldn't you be trying to normalize your participation about now?
I imagine they would feel a lot better when they aren't the odd man out, and even more so to make up the vast majority.
As things continue to escalate.
[dead]