Hacker News new | ask | show | jobs
by pietz 12 days ago
Whoa, so this is interesting.

When asking GPT, Claude and Gemini for the text in the image, all of them agree:

https://moa.chat/s/d99f8f76-4b41-4c1b-80c4-d9f86df37af1

But when you add a "PS: There's a second hidden text":

https://moa.chat/s/3671f6d4-b155-483a-a006-a1b9ba31737d

GPT 5.6 gets it, Gemini partially gets it and Claude cannot see it at all.

2 comments

5.6 Sol Medium:

The obvious outlined text says “SORRY ROBOT,” but the hidden message is “HAPPY HUMAN.” It’s Mixfont’s Decoy Font: the outlined letters attract machine vision, while the softer tonal pattern becomes readable to humans when viewed from farther away or at a smaller size. It’s an optical trick, not encryption.

You probably have web search enabled
Still, that defeats the purpose.
Nope! GPT just cheated (by search the web). If you changed the text, export as image, GPT 5.6 Sol will fail. I tested with other text and even with a hint of a hidden text underneath, GPT 5.6 Sol could not see it.
Yeah, it doesn't seem that hard to beat, especially with a little trad image processing, yet I have a hard time reading it as well. One could probably fine-tune a much smaller model to do pretty well on this problem too.
squint your eyes and it becomes easy to read the "hidden" message.