Well that worked beautifully, huh?

Jul 21, 2026 11:53 AM

TheMoonBnuuy

Views

22579

Likes

500

Dislikes

13

Providing this to show that I'm not the one who fed it to ChatGPT.

Full disclosure: I'm against generative AI and the mad dash for more compute that AI firms are doing that's destroying the environment and ruining the economy.

I'm posting this mainly to educate people that anti-AI tools are very often just snake oil that don't actually work. If something works at tricking AI, it's only the AI they tested it on, and all bets are off with different models or other versions of the same model.

They should name it Catullus, because it obviously says Sappy Roman

1 week ago | Likes 8 Dislikes 0

Sorry Robot and Happy Human I see.

1 week ago | Likes 1 Dislikes 0

But I see both, I'm a schrödinger human robot . . .

1 week ago | Likes 1 Dislikes 0

I read it as sappy human. I wonder if this font could also be used as a psychological test

1 week ago | Likes 27 Dislikes 1

i had to do the magic eye book squints to see it.

1 week ago | Likes 1 Dislikes 0

Finally a font that is improved by removing my glasses.

1 week ago | Likes 21 Dislikes 0

(removes glasses, looses sight of monitor completely)
Yeah nah, not or me.

1 week ago | Likes 3 Dislikes 0

Lol. I just tested and, yeah, same. Flawless font.

1 week ago | Likes 5 Dislikes 0

Oh god, I’m AI.

1 week ago | Likes 3 Dislikes 1

squint and you can read the other text

1 week ago | Likes 1 Dislikes 0

*Squints*
Horny Roman?

1 week ago | Likes 3 Dislikes 0

Which phrase was supposed to be easier for people to read?

1 week ago | Likes 47 Dislikes 2

Like in the thumbnail I see happy human but in the full blown up picture I see sorry robot.

1 week ago | Likes 4 Dislikes 1

The blurry one. I guess you're only meant to read it at a distance or at small scales?

1 week ago | Likes 11 Dislikes 1

Big dick in little bros bum

1 week ago | Likes 6 Dislikes 1

Cat*

1 week ago | Likes 9 Dislikes 0

...which word was supposed to be "cat"?!

1 week ago | Likes 2 Dislikes 0

Text that momentarily fooled robots and gives humans eyestrain is more like it.

1 week ago | Likes 1 Dislikes 0

Oh great another font to use in captchas.

1 week ago | Likes 7 Dislikes 1

So those cringe meta glasses *will* have a use?

1 week ago | Likes 2 Dislikes 0

1 week ago | Likes 3 Dislikes 0

oll sllit

1 week ago | Likes 1 Dislikes 0

How would the TTF work? If you are to hide a message, the 2nd message needs to make a little bit of sense. So you'd need 26*25 letter characters (650) and I'm ignoring symbols and punctuation. I think it's even worse than snake oil.

1 week ago | Likes 1 Dislikes 0

I think it's more a system than anything else

1 week ago | Likes 1 Dislikes 0

In the post it says "Decoy Font is available in TTF format and can be used like a regular font." but there is no way to write 2 messages in one with a regular font.

1 week ago | Likes 1 Dislikes 0

that's what it says, and I agree with you that there's no way to do that, which is why I say it must be a system...
it's likely two fonts with the same kerning

1 week ago | Likes 1 Dislikes 0

Probably. I'm definitely being pedantic here, but if we need a photoshop-like app to use 2 superposed fonts then "can be used as a regular font" is a stretch.

1 week ago | Likes 1 Dislikes 0

well if you think about it, if it's not baked into in an image, it's not going to stop an AI that reads text. LLMs don't see fonts... I mean like [font=decoy, size=14]blah blah blah[/font] isn't exactly going to conceal the "blah blah blah" from the AI.
This is likely meant for making signs for the real world so computer vision AIs can't read it, but that's unlikely to work, as the cameras are more likely to pick up the blobby text than the thin lines.

1 week ago | Likes 1 Dislikes 0

Fuck.....Im a robot

1 week ago | Likes 1 Dislikes 0

Totally understandable: FURRY PORN

1 week ago | Likes 2 Dislikes 1

Whithout posting the descriptive text, ChatGPT can't see the HAPPY HUMAN. Give it a week, I'm sure it will.

1 week ago | Likes 2 Dislikes 1

It works reasonably well. If you don't prompt the bot to look for double meaning then it works out the gate. If that ever becomes part of a standard prompt template then it more than doubles the compute costs of scraping which will protect some content from being scraped.

1 week ago | Likes 2 Dislikes 1

Should I be worried that I saw it as Sorry Robot first and had to struggle to see Happy Human?

1 week ago | Likes 1 Dislikes 0

No. It's intended that you have to squint to see the human only text

1 week ago | Likes 1 Dislikes 0

One of the most frustrating things I've gotten is that I've been a professional writer (in one capacity or another) for 20+ years. AI was trained on the *best* professional writing.

And when I post my original work, I often get accused of writing "AI Slop." I literally have to make my writing *worse,* deliberately, for it to be acknowledged as *my work.*

1 week ago | Likes 12 Dislikes 1

Yeah I see that a lot. Certain styles of art are instantly accused of being AI. Certain writing or design is instantly accused of AI. I've seen artists I've been semi-following for over a decade get accused of AI since their art style is vaguely similar. It's super weird watching everyone accuse anything that's "not good enough" or "too perfect" as AI. Budding artists need to either be terrible or really-good-but-not-at-these-styles to not get scrutinized.

1 week ago | Likes 3 Dislikes 1

And I get it. Some people are cranking out AI anything and claiming to have made it, You feel like you gotta scrutinize everything or get tricked. But man it must suck to be a budding artist who kinda sucks at drawing hands.

1 week ago | Likes 1 Dislikes 1

Fun fact: AI detection tools are themselves AI that are trained to spot differences between human output and AI output, with AIs being trained specifically to produce human like output.
They BARELY work with modern models, and when it comes to artwork, they basically don't work at all because the main market is detecting AI generated photos because that's what the general public is concerned about: is this photo of my favorite politician AI? and things like that.

1 week ago | Likes 4 Dislikes 1

And since artwork never looks like a real photo, AI image detectors almost always flag all artwork as AI made

1 week ago | Likes 1 Dislikes 1

As a person struggling to chose even between O and 0, I say 4@єk Voи

1 week ago | Likes 1 Dislikes 0

the more shit it has to learn the harder it is to be good at what it does.

1 week ago | Likes 1 Dislikes 0

This timeline is so fucking dumb

1 week ago | Likes 92 Dislikes 1

I want the timeline where the greedy megalomaniacs burning the country down for their own profit have scales, wings, horns, and nobody tries to accuse you of murder when you hunt them down on horseback.

Our dragons suck. >:(

1 week ago | Likes 8 Dislikes 0

You suck for advocating the murder of innocent Dragons!

1 week ago | Likes 2 Dislikes 0

fun fact: all observations point to this being the most probable time line...
meaning either all the other timelines are even dumber, or less stupidity is improbable

1 week ago | Likes 6 Dislikes 1

Wha... That wasn't fun at all!

1 week ago | Likes 1 Dislikes 0

What? No, that's not how probability works. Improbable things happen all the time, just not usually. There are MANY nexus points in history that are highly chaotic and sensitive and the outcome appears to be largely left to chance. The odds of the sun rising this morning were pretty good while the winner of the 2004 Democratic primary really could have gone a bunch of different ways. Just because this is what happened, doesn't make it the most probable.

1 week ago | Likes 3 Dislikes 1

NOT what I said. I said this timeline is the most probable timeline, NOT "this timeline consists exclusively of only the most probable events". A timeline consisting ONLY of the most probable events is actually ironically extremely improbable.
Take a 20 sided die. Label every face except one "1". The one face that isn't 1 is "2".
1 is the most probable outcome. But if you rolled it a billion times and only ever got 1, that'd be highly improbable, now wouldn't it?

1 week ago | Likes 3 Dislikes 0

" I said this timeline is the most probable timeline," And that is exactly what I was responding to. That is not how probability works, pretty much for exactly the explanation you yourself have stated.

And you're just kinda making stuff up. WHAT observations?

1 week ago | Likes 1 Dislikes 0

No, you interpreted "most probable" as "every event only went with the maximally probable outcome"
I then clarified the matter for you, and you're now arguing with me about what I meant by that. Since you're allowed to make up what I meant and I can't clarify what I was saying to correct your misinterpretation of that, it is now clear that you are only here to argue in bad faith.
So there is nothing of value to be had here, as value can only be had from good faith conversations.
Goodbye.

1 week ago | Likes 1 Dislikes 0

So how long until this font is outlawed

1 week ago | Likes 1 Dislikes 0

Harry Roman?

1 week ago | Likes 1 Dislikes 0

Anything that can trick AI or that is used to identify AI stops being useful as soon as the people in charge of AI hear about it because those are the things they will focus on. Even the hands keeps getting better.

Just like with finance, a search engine, social media manipulations, or even business reviews, the only way to keep it working is to keep the "test" a secret.

1 week ago | Likes 2 Dislikes 1

well here's the thing... it's usually not even effective on the AIs run by large AI firms to begin with.
See, the anti-AI tools are created by testing against the open source models you find on HuggingFace and similar repository sites. The models you find there are made to be downloaded and run at home on home computer, albeit beefy home computers.
The commercial AIs like ChatGPT, Claude, Grok, etc are all black boxes. We can't test if training data poisoning works on them.

1 week ago | Likes 2 Dislikes 1

So at BEST the anti-AI tools will stop home labbers from using your stuff... except... it doesn't work because like with images you can just apply a 1% gaussian filter and BAM, poison defeated.
So not only does it NOT work on the huge data scrapers, it doesn't even really work on the AIs it's supposed to work on.

1 week ago | Likes 2 Dislikes 1

Poison can work if it's widespread enough... but, like, it needs to be excessively widespread. Widespread enough to the point of (possibly) diminishing all of our own web browsing experience.

1 week ago | Likes 1 Dislikes 1

with my one bionic eye sorpy hubot

1 week ago | Likes 1 Dislikes 0

I keep seeing Harry Roman

1 week ago | Likes 1 Dislikes 0

While scrolling usersub, the thumbnail looked like it said horny human.

1 week ago | Likes 174 Dislikes 4

Then it fits right in with this site!

1 week ago | Likes 1 Dislikes 0

Sigmund Freud would have a thing or two to say about that 😂

1 week ago | Likes 24 Dislikes 0

I heard that a Freudian slip is when you say one thing but really mean your mother.

1 week ago | Likes 5 Dislikes 0

The most accurate reading of all.

1 week ago | Likes 7 Dislikes 0

There's a secret, third way to read that font.



For us degenerates.

1 week ago | Likes 3 Dislikes 0

That's the one thing robots can't feel. Checkmate.

1 week ago | Likes 2 Dislikes 0

1 week ago | Likes 2 Dislikes 0

Well I'm glad I wasn't the only one.

1 week ago | Likes 2 Dislikes 0

Internet has ruined us...

1 week ago | Likes 2 Dislikes 0

I thought it said horny roman at first

1 week ago | Likes 1 Dislikes 0

I saw Harry human

1 week ago | Likes 2 Dislikes 0

Yer a human, Harry.

1 week ago | Likes 2 Dislikes 0

I read HAPPY HYMEN

1 week ago | Likes 2 Dislikes 0