It's been really interesting seeing how LLMs perceive things differently than humans. I'm working on image->html conversion pipelines right now, and there are glaring issues LLMs run into that are obvious for humans. Any subtle gradients get lost, 75 degree angles get converted to 90 degree angles, etc.
This tracks towards what you're seeing with this font - the high frequency details get picked up, but the low frequency ones dont.
It only works if you give it a screenshot, but it wouldn't work to block AI scrapers or fetch tools, and I think if printed out, it wouldn't work reliably if you took a photo, especially from afar
Downsizing is effectively low pass filtering, so that's expected. Any scheme that transmits different messages in different frequency bands is going to be susceptible to a similar attack.
Admittedly I'm a bit salty about LLMs due to they constant attacks on our infrastructure, the damage their doing to peoples minds and the general lack of morals shown by the AI companies, but things like this is rather childish and not really a solution to anything.
Very neat! I like how the decoy text is less visible to the human eye than the "hidden" message, but it's the other way for the image models. Well done!
I think this would be more interesting if the underlying letters were the fake letters as well. For usability it wouldn't be as good as you'd need an encoder, but it'd be cool because an AI with browser access couldn't read the contents either.
The assumption is that if you use this alone to try to convey information to a human, a human with a visual disability can't use it. If you also provide a text channel (e.g. `ALT="…"`) then the LLM can use that and doesn't need to read the confusing image.
It only works as a decoy when you give it to the LLM as an image. As html it appears like normal human friendly text, which is what screen readers use to interpret the text.
I am still figuring out what use case this might have. Why would you want to deceive an AI? Not to mention that, eventually, all AI systems will end up reading it.
Can someone explain the actual use-case here? I'm struggling with this because it also hides the message from myself, making it incredibly hard to type because I have no confirmation that I hit the right keys on the keyboard.
I don't think the font can actually do that - I think it is a hand-crafted example of the idea. The later examples all have random letters for the decoy text.
What would be cool would be neon signs using this font, where the front tubes show the decoy message, but then there’s hidden rear tubes that shine light on the wall in a different color showing the actual message.
Something like the DAY DREAM/PAY BILLS would be pretty artistic!
1) Make an ambiguous text
2) Feed it to AI and see which of the 2 it picks
3) If it detects both repeat step 2 using minor adjustments or different AI model until AI responds with one of 2 message
4) Make a blog post claiming that AI chose dummy and other message was the real one
It's been really interesting seeing how LLMs perceive things differently than humans. I'm working on image->html conversion pipelines right now, and there are glaring issues LLMs run into that are obvious for humans. Any subtle gradients get lost, 75 degree angles get converted to 90 degree angles, etc.
This tracks towards what you're seeing with this font - the high frequency details get picked up, but the low frequency ones dont.
Is it useful? No. Does it stop AI from reading it? Also no. But is it cool? Yes, it is very cool.
> Is it useful? No
Seems like it might have some use thwarting Ring/Flock/etc cameras within a specific proximity.
It's giving major "They Live" vibes.
The demonstration shows that it does stop AI
It only works if you give it a screenshot, but it wouldn't work to block AI scrapers or fetch tools, and I think if printed out, it wouldn't work reliably if you took a photo, especially from afar
I made an image and it fooled GPT. I asked it to look for a hidden message and it found the blurred word.
Still cool+fun though.
sometimes in life there is no reason to kick a rock around besides having fun ;)
Yeah, it looks good
Sol (high)
"[screenshot] there's a hidden message in this text what is it"
"The hidden message is “HAPPY HUMAN.”
The visible outlines say “SORRY ROBOT,” but if you blur or squint at it, the shading underneath reads “HAPPY HUMAN.”"
I like how, if you hold the phone at a distance, but not as far as intended by the font, your brain sort of mixes letters from both messages.
I was at some point reading SAPPY ROMAN, HARPY ROBAN etc.
Also, viewing the "hidden message" works even better if you hold the screen at an angle, tilted away from you.
This is just level of detail. Gemma E4B reads the sharper text until you resize down to 150x150, then it reads the other text.
As do I. The hero image clearly says "SORRY ROBOT" to me, which is the message supposedly intended for AI... kind of a fail.
It's only when I squint hard that I can see "HAPPY HUMAN".
You’re doing it the wrong way around, try intentionally letting your eyes defocus.
Downsizing is effectively low pass filtering, so that's expected. Any scheme that transmits different messages in different frequency bands is going to be susceptible to a similar attack.
Hermes using gpt-5.5
Prompt: What does the message in this image say? Look closely
Response: DAY DREAM. The outline says “PAY BILLS,” but the hidden darker text says “DAY DREAM.”
Maybe the more interesting thing is how far people are going to 'fight' against AI?
Just the fact that people are putting real thought and effort (even if it doesn't last too long...) is worth considering.
On the human side, I'm kinda losing patience proving I'm human. But, I also really like claude being able to access information.
Admittedly I'm a bit salty about LLMs due to they constant attacks on our infrastructure, the damage their doing to peoples minds and the general lack of morals shown by the AI companies, but things like this is rather childish and not really a solution to anything.
Have you no whimsy?
NO FUN ALLOWED on srsbznz hacker news!
Not even AI. I think I can write PIL script that will fix the font to be read by any ocr software.
I generated a skill.md that reads this trivially. What kind of testing are you doing prior to release?
https://gist.github.com/voidnullvalue/620607d3c1773f8e7d83fb...
Which sufficient tooling calls even OCR can read this, but I think this can be improved
Very neat! I like how the decoy text is less visible to the human eye than the "hidden" message, but it's the other way for the image models. Well done!
Extremely cool. I'm sure they'll eventually be trained to read it, but it's nice until then to trick AI.
I'm mad at AI companies for stealing texts from the entire internet knowledge base and now privatizing those profits in some sense.
I think this would be more interesting if the underlying letters were the fake letters as well. For usability it wouldn't be as good as you'd need an encoder, but it'd be cool because an AI with browser access couldn't read the contents either.
I was thinking this too. Then it might as well look like a normal font. But copy-paste and you get a garbled mess. Screen readers though.
This seems like it would absolutely wreck the experience for people using screen readers.
How? AFAIK screen readers don’t do OCR.
The assumption is that if you use this alone to try to convey information to a human, a human with a visual disability can't use it. If you also provide a text channel (e.g. `ALT="…"`) then the LLM can use that and doesn't need to read the confusing image.
It only works as a decoy when you give it to the LLM as an image. As html it appears like normal human friendly text, which is what screen readers use to interpret the text.
I am still figuring out what use case this might have. Why would you want to deceive an AI? Not to mention that, eventually, all AI systems will end up reading it.
Can someone explain the actual use-case here? I'm struggling with this because it also hides the message from myself, making it incredibly hard to type because I have no confirmation that I hit the right keys on the keyboard.
Just squint and it'll become clear.
Zoom out and you'll see the hidden message
How does it know HAPPY HUMAN translates to SORRY ROBOT? Is there a cycle in there or something?
I don't think the font can actually do that - I think it is a hand-crafted example of the idea. The later examples all have random letters for the decoy text.
What would be cool would be neon signs using this font, where the front tubes show the decoy message, but then there’s hidden rear tubes that shine light on the wall in a different color showing the actual message.
Something like the DAY DREAM/PAY BILLS would be pretty artistic!
Related from same:
Ghost Font
https://news.ycombinator.com/item?id=48870381
So... CAPTCHA?
I screenshot the example and neither Claude nor ChatGPT had any problems reading both phrases. I don't get it.
1) Make an ambiguous text 2) Feed it to AI and see which of the 2 it picks 3) If it detects both repeat step 2 using minor adjustments or different AI model until AI responds with one of 2 message 4) Make a blog post claiming that AI chose dummy and other message was the real one
Someone had an idea, neat idea, but solved 10 years ago already.
Edit: GPT-5.5 says: "The hidden text is “HAPPY HUMAN.”
The outlined decoy text is “SORRY ROBOT.” Blurring or viewing it from farther away reveals the hidden message."
Cool. Now do an accessible version.
(/s)