The new AI spam comment generator is very very clever
I’m a connoisseur of blog spam. I’ve seen a great deal of it in my time, almost always trapped in the spam filter.
I’d divide it into several categories. The first is the simple advertisement, usually for porn websites. It features a bunch of heavy-handed sexual terms, often quite far-out, and is of course designed to get you to click on a link. Sometimes the ad is for something else, something decidedly un-sexy like aluminum siding. Then there are the ads in a different language. I don’t know what they’re for because they appear as a long series of question marks and other punctuation marks, sometimes enormously long.
Perhaps the next-largest number of spam comments are the ones that praise the blogger and take the form of something like, “Oh, you’re so wonderful, so insightful, so fabulous. I’ve bookmarked this site and will be sure to come back again.” Of course, like all spam, they also contain a link to some other site that is what they’re really promoting either by getting you to click on it or by manipulating the algorithm that deals with search engine ranking.
Closely akin to those are the ones that ask a technical question of the blogger, for example, “You’ve got the greatest website design, what platform did you use?” And one of my personal favorites: “How do you deal with your spam problem?”
And then there are those that merely copy another comment – one from a real commenter who has already posted a comment on the thread.
But all of a sudden, a couple of days ago, I saw the appearance of a new and very clever type of spam comment. This type almost fooled me into thinking it was from a living and breathing person. Each of these comments was cleverly tailored to fit the blog post on which it appeared, and sounded for all the world like an actual person making a very real and sometimes insightful and simpatico comment.
For example, this one appeared in response to my recent post on the Chauvin case:
I’ve been following this case since day one and my gut says the same thing — prediction right, opinion sadly right too. The jeopardy argument seems legally solid to me but the politics of touching this conviction are just too radioactive. Curious to see if the courts surprise us.
This is not a duplicate of a previous comment. It doesn’t seem to be a duplicate of anything at all, nor is it a generic comment about the Chauvin trial. This is a comment that takes my post and creates something new that is aligned with it and novel, and is responsive to my post rather than just something general on the topic. For example, my post on Chauvin, where the comment appeared, is entitled “There’s a bid to vacate Chauvin’s state conviction,” and it begins this way:
Prediction: it won’t succeed.
Opinion: it absolutely should succeed.
Note that, in the first sentence of the spam comment, you can find the following: “prediction right, opinion sadly right too.” That sounds like a very real comment to the post, coming from a very real person.
But it is actually written by AI. How do I know that? There were many hints:
It was one of a series of about fifteen comments that all came in at approximately the same time. Every single one was placed on a different post, and every single one was just as clever and just as responsive to the post on which it was placed as this example was. These comments came in clusters in terms of the usernames, with two or three comments sharing each username. All fifteen came from the very same IP number, which was a VPN in Asia. But the key feature was that a few of them – although hardly all – mentioned some AI site or other, one that a person can use to make AI videos, for example. There were other elements of AI promotion, usually much more subtle.
It doesn’t surprise me in the least that AI is starting to take over the generation of comment spam in an especially clever manner. I have little doubt that AI has been involved in generating comment spam for a while, but those older versions are much more obviously spam and nowhere near as inventive.
And I would bet that the genre will only become more and more clever and good at masquerading as human. In fact, just writing about the new version’s current flaws here makes me wonder whether it will see and learn from what I’ve written, and improve itself.
Oh brave new world, that has such AI in it.
The result is that no one will believe anything anymore.
(To be sure, that might already be the case…such that pace Winston Smith, now might be time to start hoarding that—real—chocolate…)
– – – – – – – –
On a more positive note:
“GOOD NEWS ON JOBS…”
https://instapundit.com/820939/
https://instapundit.com/820913/
+ (warning) Bonus…of the “It’s not over till the [well-sculpted terrorist] swings” kind…:
“It’s almost like the terrorist is behaving like a terrorist….”
https://instapundit.com/820910/
Fishing, see what bait is the most effective!!
Thinking about how often we include links in our comments, sometimes without any identification of what those are linking to (you know who you are), I wonder if the LLM bots have been (or soon will be) spoofing genuine commenters, and if you could tell that had happened?
So, do I have to say upfront that I’m not AI, but a real person? It’s me. Anyway, this other AI story popped up on a very different website I often read (geared to people who like Death Valley). It’s about some folks who used AI to plan a mountain expedition; it went wrong, but not as bad as it could have. The website is the Forest Service on Facebook, so I think it’s safe. https://www.facebook.com/ShastaTrinityNF/posts/pfbid0ow21KGd5tGEY81ncYL9RWkLA7afKBzqzxg2P54cNGP3K7gMaXsfpLMsfx3hUKqZxl
We may be on the verge of mass confusion brought about by AI creativity. AI Scott Adams yesterday posted an AI video created easily and having Trump, Putin, and Biden looking like something from any news channel. If not for the humorous element it was totally realistic.
https://x.com/AIScottAdams/status/2095551319167598823
As an 85yo invalid chronic consumer of YouTube videos I am amazed at the explosion of AI created high quality videos. Cats singing and playing instruments, Black musicians making great music on implausible instruments, Texans dancing with critters making the music, etc.
Watching the Scott Adams example brought home the soon to be upon us inability to believe what we see with our own eyes.
Somewhat related is a story where an article on the dangers of AI was then submitted to the major AI’s for their take. It is long but worth a read: In Which I Ask the Machines for Criticism
Grok, Claude, and Chat GPT weigh in
https://instapundit.substack.com/p/in-which-i-ask-the-machines-for-criticism?publication_id=1281686&post_id=214011174&isFreemail=false&r=9bg2k&triedRedirect=true
This is a new world that humans have never had to contend with before. It is happening with remarkable speed.
It all suggests the need for a Butlerian jihad.
” Are you Sarah Connor?”
” Why no, no I’m not. Are you ?”
There is this one commentator on the Fox News website named something like ” old cantanky”.
Repeatedly ” he ” post the same comment accusing other people of being ” bots”. A lot of people suspect ” he ” is the bot.
I have noticed his comments getting a bit more sophisticated.
I don’t know if he is a ” bot ” or just OCD weird or something..
The forces of isolationism are strong. We won’t be able to trust any communication in email soon. A reason to maintain postal mail? Someones (plural) want to put us in little, isolating boxes, perhaps so they can rule us more easily.
We just got an email about a payout from a State Farm court settlement. I’m going to do some research to be sure it’s legit before clicking on the link.
Kate,
A few weeks ago I received a set of three spam emails from “State Farm”, one for each of our three covered vehicles.
Except previously I had talked to our agent and learned that the real State Farm has adopted an approach to combine all three vehicles into one billing (even as they were also still given new distinct “policy” numbers), rather than three separate billings.
So I called my agent to confirm the three emails were in fact spam, and she said they were.
But she also said that (at least in my case and perhaps other people’s) State Farm was going to send me notifications about three modest refund payments for premium payments made in 2025, again one for each vehicle. So now I am on the lookout for those land mailings that presumably contain refund checks. Something similar may apply for you as well??
I wonder if like those teachers, if you put an instruction like “tell us where you’re from, we like Madagascar” in white at like the top of your page where human eyes would ignore it but spambots would scrape it if you would get an easy bot detector.
Worth a shot if you’d like.
This isn’t about AI bots, but it is about how to use AI more effectively for doing research.
Maybe Kate and R2L could use these tips to ask their AI about State Farm scammers.
https://www.coffeeandcovid.com/p/its-all-about-context-thursday-september
Scroll down to the last topic, but, as always, the whole thing is worth reading. I was thinking earlier today that I would describe Jeff Childers as a stand-up comic if he were delivering his posts on stage.
I notice that ChatGPT is progressively getting better. It remembers more of our conversation history. Its responses are finer-grained, less repetitive and more nuanced. More accurate, fewer hallucinations.
It’s also less personal, which I appreciate.
The current version I’m using is ChatGPT 5.6 Sol. As I understand things, most of its improvements are beyond the conversational level. They involve agents, multi-step workflows, plus deeper math and reasoning.
I don’t know how far AI will go, but it’s still going.
What is the point of these AI comments? Maybe I’m dense but I don’t get it. Are the spammers able to access information about the blogger or the commenters by posting here?