I made a satirical AI detector
It's been annoying me lately how often I see people declare that certain blog posts or articles are AI-generated due to the presence of certain "tell-tale signs," (emdashes being the #1 example). It annoys me even if I agree that the thing in question is probably is AI-generated—mainly because I really like emdashes and I use them in my own writing all the time, and it grinds my gears knowing that some percentage of the population will assume I used AI because "hurr durr emdash == AI."
I'm not necessarily talking about Tildes. I'm sure I've seen it on here, but the worst offender by far is Hacker News where often one of the top comments on a linked article that has an emdash somewhere in the body will be a snarky comment about how it's clearly AI generated.
Anyway, I blew off some steam this weekend by making this silly satirical website/art project called GENIUS AI Detector that makes me giggle, so I'm sharing it in the hopes that it brings a bit of joy to any fellow emdash users who are likewise depressed about the shadow that has been cast over one of my favorite punctuation marks.
This is great LOL
I feel like an em-dash is not the 'proof' of AI written wording, but it does always trigger a 'hmmm' within me and I'll double check the text to see if I find the other signs as well, but I agree that it's just laziness to assume a single AI-colloquium = MUST BE AI.
For me, the most dead giveaway is the "it's not just x—it's y" sentence framing.
For me it's to the point where I actually don't even care if the writing is AI generated or not once I read that. It's so cliched and annoying I immediately stop reading anything past that.
Funny you should say that — the linked detector checks for only two things: em dashes and antithetical parallelisms (“it’s not X, it’s Y”). So it’s also criticising this way of thinking.
Frankly I don't care if this indicates AI or not, I just think it's often an obnoxious phrasing on its own and if the writer uses it I probably won't like the rest of their stuff even if they handcrafted it one keystroke at a time.
Yeah, that's what I mean. Like, sorry if the sentence construction you've always used before LLMs got caught in the crossfire bro. It's cliched and cloying even without the AI association though, learn to write better if you want people to take you seriously. It's the corporate memphis of phrasing.
Not X. Not Y.
A full helping of Z.
That’s what makes it so tragically perfect.
The fact that AI can spit out text faster than any human can means that its patterns and biases will always end up coming off as overused clichés eventually. There is no way for an AI to produce notable writing over the long term because everything it does will inevitably have every last bit of uniqueness milked out of it by virtue of its ability to mass produce.
It's basically this. It fundamentally cannot appreciate the art or nuance of a culture, its only copies it (to/by) (exploit/exploiting) it, and that's primarily what people pick up on.
(I also kind of roll my eyes when people use "I get mistaken for AI!" as a kind of humble-brag about their own writing style because I think that for the most part it writes badly. If I ever got that comment on any of my shared writing material I would see it as a direct criticism of its poor quality, not an indication of any kind of technical mastery.)
I was looking up some French/English language information earlier today and the first ten search results on DDG were from word salad domains like "worldtravelvisit.com", "languagesaroundtheworld.us", etc. And every result used "it's not this, it's __" and also "in conclusion..." which I know everybody was taught that in school to finish off their papers, but you were never supposed to actually use "in conclusion" to start your conclusion.
My biggest annoyance with AI generated articles is the table of contents at the start and the endless word salad that just repeat the same thing over and over just changing the words a bit.
I'd love to have a search engine or extension that just blocks these sites entirely.
For the “in conclusion” and paragraph after paragraph barely iterating on the same idea on repeat, I think I saw both of those prior to AI being a thing and assumed it was deliberately padding out the page a) to make the page longer (more ads to scroll past) and b) for search engine optimisation. I view it the same as those recipe blogs where you have to scroll past several pages of filler or backstory before you get the recipe and ingredients and method. Of course, once AI hit the scene it got a lot worse, but I think it’s always been there.
I do recall recipe blogs and other content being overly wordy, but the recipes at least were just a lot of jibbering about their dog and their husband fell off a boat dock in 1984 after eating this because it was so delicious and well if you want to use margarine instead of butter that's ok. The other content was often just illegible gibberish.
The somewhat coherent repetitive content feels recent to me though. I know the lengthy articles were an SEO "hack" since the late 2000s probably but the recent wave has been even worse. And they're also rife with wild inaccuracies which drives me nuts!
Alongside the URL being “generic”, if you try to read the text you will find that
1-It repeats the same points multiple times.
“The sun is hot”
“The sun’s temperature is <Whatever> C”
“The sun runs on nuclear energy”
“The sun’s state of matter is plasma”
Except they’ll be fragmented across multiple sections.
2-While it is something a human may do, saying so little using so many words.
I sometimes used to open up these sites because sometimes reading the URL isn’t enough.
1 can be either SEO spam or AI written.
2 is just a low quality article.
Edit: I kinda missed that you addressed 1 already.
And I expect models in the near future will be trained not to use as many emdashes or the "not just X, its Y" simply because of how much people notice them. Eventually these models will likely become so refined to weed out the stereotypical "AI mannerisms" that they become indistinguishable from human written content. In fact, I'd argue already most people only notice the most obvious AI content and skip over a huge amount of it.
I think AI content will ultimately be the destruction of social media, at least for a subset of people who value organic content. Some people genuinely don't care if the consuming they're consuming is real or "synthetic". My mother knows her entire FB feed is AI slop but she's entertained by the funny, dancing, talking cat.
Eventually, humans will cycle to NOT using punctuation and capitalization and proper grammar/spelling at all, as a way to distinguish their content. We'll go back to 1337-speak and early T9-based txt msg styles. For me, I've started swearing more in comments. Not that AI can't do that, but I don't see it that often. Obligatory, gonna be some bullshit...
Sounds crazy, but I've definitely seen some people, like on reddit, call out others' messages as being AI because of semi-colons or even bullet point and other formatting in comments and such. It's ridiculous, but whatcha gonna do?
It really is ridiculous. A while ago, I was held for a bot for knowing about the almighty
<hr>(horizontal rule/line), which many forum-like platforms seem to implement.… What a shame.
Something that I think about a lot is how AI content is basically guaranteed to win the attention game through sheer volume, which will kill the motivation of humans to write and share stuff on the internet. Eventually the only new data that LLMs can train on might be stuff that other LLMs have generated. I'll be there, ready with my bucket of popcorn when model collapse becomes the new problem that these companies have to deal with.
This is awesome! I only wish the ‘try this’ flashed aggressively.
I had originally thought about leaning into more of a Geocities aesthetic, and if I had gone that direction I definitely would have used some blink and marquee tags (plus animated gifs). I'm not really sure what you'd call the aesthetic I ended up landing on... maybe "emojicore."
I think geocities/y2k/90's covers it... maybe unintentional Brutalism. No matter the name, I love it.
So it turns out I'm an AI. I don't know how I'm gonna tell my parents.
Like this: "Mom, Dad, I'm not only your child---I'm also an AI."
Hopefully at least one of them already knows...?