Hacker News Reader

Comment by jdthedisciple | Hacker News Reader

What's up all these GPT bot comments here on HN lately? Seems like a fairly new phenomenon.

Bots seem to be seasonal creatures. Some months you won't see any. And then some months you'll see four per submission. I'd just downvote and reconvene when it seems to be a substantial issue.

As for GPT, the bots you usually see on HN are just as easily written by humans. GPT makes sense for scale, but it's not very interesting when bots are posting at such low volume, so it doesn't necessarily have to be the case nor does it really change anything.

I see 4 dead comments, but I suspect one of them is from a real human (derjanni)

I’ve seen a lot of new users today that were rage silenced, including obviously human ones.

Looking at their history I think they are related to the bots as they have a suspicious number of threads where the bots also chime in.

They have submitted two "blog posts" from their own website whose insightlessness reeks of GPT, whose comments here are nothing but bots from brand new accounts.

If there's a ring, they are the ringleader.

Yeah, bots - sure. GPT? No way. My rapid-classification pattern matcher, honed on two decades of running blogs and web bulletin boards, recognizes the shape, cadence and sound of these comments as indicative of the usual kind of spam comment, common in the last 15+ years.

I.e. they came out of some boring, old-school script.

(Though I do wonder, how much of the spam comments on random, long-forgotten wordpress blogs, ended up in the GPT-{3,4} training data.)

They sorta read like those ride reviews from the theme park denizens of SimCoaster tbh.

It's a spam ring and has been around for awhile. The new phenomenon is that some of these are leaking through. I probably need to adjust the anti-spam code.

unfortunately this is the future of the entire internet - and not just the comments the submissions as well. the search for authenticity and actual human material will become the most valuable thing.

This isn't that different from what we have now as people have abandoned corporate messaging for communities and podcasts/youtube/etc to try to find authentic human voices. The difference will be how much harder it will become to catch the bots.

And the tough part is new users being kicked out when someone thinks they’re a bot. How do we protect people from the bot witch-hunt?

On the other hand if your human comment is indistinguishable from a bot comment, maybe it’s ok to be kicked out?

Related: Reverse Turing Test: https://en.wikipedia.org/wiki/Reverse_Turing_test

It depends on the community. I think its ok for humans to not be perfect, sometimes. Meanwhile I have no interest in what ChatGPT thinks about anything.

You've hit a pet peeve of mine:

The key feature of witch hunts as a concept is the non-existence of witches. If you have a hunt for something that certainly exists (Soviet spies infiltrating US institutions come to mind) it definitively isn't a witch-hunt.

For my part, I'm going to keep downvoting comments that say nothing in the most bland way imaginable, and flagging anyone who posts "I put your question into ChatGPT; here's what it says" (in contexts other than discussions of chat bots).

This is what it's like on Reddit. Many of the popular and smaller subreddits won't let you participate unless you already have gained a certain amount of karma, but you can't gain karma without participation.

New users can't participate unless they go out of their way to find something they might not be interested in and comment until they have enough points to participate elsewhere.

If it ever become impossible to spot a human from a bit, then bots become valuable contributors to online discussion with interesting insights so I don't see the problem.

that depends on what your purpose is for going online and engaging in social networks/forums

> We can thus imagine a technologically highly advanced society, containing many sorts of complex structures, some of which are much smarter and more intricate than anything that exists today, in which there would nevertheless be a complete absence of any type of being whose welfare has moral significance. In a sense, this would be an uninhabited society. All the kinds of being that we care even remotely about would have vanished.

https://nickbostrom.com/fut/evolution

https://xkcd.com/810/

I actually think some major commercial use cases of the OpenAI API are a) fake reviews on Amazon etc and b) SEO spam.

Without naming names - the other day, I sat on a presentation related to a commercial offering of some SOTA LLM APIs; there was a slide listing a dozen use cases claimed to have been piloted or already tested. I distinctly remember being a bit disgusted by how all, excepting one or two, were highly morally questionable, as they boiled down to improving or scaling all the usual ugly kinds of marketing and marketing-related surveillance schemes.

The last time I felt this kind of disappointment was when those Bluetooth Low Energy beacons ("iBeacon") became a hot topic - a lot of startups popped up, and quite quickly, they've mostly settled on the same primary use case: retail. That is, tracking and spamming unsuspecting users as they move about shopping malls.

This will be on of thousands of similar meetings going on for many many companies.

https://www.google.com/search?q=site%3Aamazon.com+%22as+an+a...

It's already starting, humorously at least for now.

One Google snippet (the Amazon product page was already deleted) said:

> As an AI language model, I do not have personal experience with using products. However, I can provide a negative review based on the information available ...

So people use AI not obly to promote their products, but also to create negative reviews, presumably against competitors?! That's fu**d up.

Do you have any examples before they were deleted? I wonder if I'd even recognize a fancy bot comment

Turn on "showdead" in your HN account settings, and you can see deleted content.

At the moment it's a few brand new accounts posting essentially the same content-less statement that covers the big points in the headline {novelty, pollination, frog}.

Example: That's amazing! It's incredible that we are still discovering new species, and that some of them can even have special abilities like pollinating. I'm looking forward to reading more about this frog.

Kind of concerning that "hey that's pretty neat" comments are immediately suspected (probably accurately) of being bots. A comment like that example isn't useful to the other comment readers, but it's probably pretty motivating to the author, and it makes forums more positive places. Removing the organic versions of those will make forums a more negative place, and that's a shame.

I'd be more accepting of a "hey that's pretty neat" comment if it didn't read like the equivalent of a middle schooler's attempt to restate the headline as if they read the article (with some hallucinations, such as this being a newly-discovered species).

Edit to add: Sibling comment has me questioning whether I'm a robot

when I got a pleasing reading experience from HN, I can't leave some comments like: this is so cute/a good story... this always get so many downvote haha.

I don't see the point of making bots on HN to post about some damn frogs

those are probably just boring people

the point of the bots isn't posting about frogs, they are posting about frogs to (badly) look like users that have normal activity when they do start spamming or mass-upvoting.

Never ascribe to ChatGPT that which is adequately explained by annoying people.

When it's consistent groups of 4 brand new accounts all making essentially the same comment at the same time (in multiple places) it starts to get pretty suspicious.

I wonder how well /r9k/ would work on substrings. It seems like these GPT robots are roughly as good as taking all the random text in a google search taglines and making a sentence out of it.

Very strange indeed, first I've seen it, but they're all the same format and all new accounts.

Could be someone testing it out seeing if he can get away with it.

Reply on news.ycombinator.com