Decoy Font
A new “decoy font” experiment overlays two messages in one image: sharp outline letters that current vision models tend to read, and a blurred background word that humans can better perceive by squinting or viewing from a distance. Commenters see it mainly as a clever optical illusion or art project rather than a durable anti‑AI technique, noting that models can already be prompted or trained to recognize both messages and that real-world use would raise usability and accessibility problems. The broader theme is how far people are willing to go to “fight” AI scraping, and whether such tactics meaningfully slow models or just make experiences worse for humans.
Concept and Mechanism
- Font overlays two messages: a sharp “decoy” text and a blurred “hidden” text.
- Humans can often see the hidden message by squinting, defocusing, zooming out, or shrinking the image; models tend to read the sharp outlines.
- Some commenters only see one of the messages depending on zoom, angle, glasses, display theme, or eyesight.
Effectiveness Against AI
- Several people report that current vision-language models (GPT, Claude, Gemini, others) can already read both messages or infer that squinting/defocusing reveals a second text.
- Others share tests where models only see the decoy (e.g., “PAY BILLS”) or misread letters, suggesting partial success.
- Some suspect models are “cheating” via web search or matching known demo images.
- Multiple comments note it’s trivial to defeat via simple image processing (downsampling, blurring, morphology filters) or by training a small model specifically on this style.
- Consensus trend: may confuse some current models in narrow conditions, but not robust and unlikely to scale as a long-term AI defense.
Usability and Accessibility
- Many find it hard or uncomfortable to read; some experience eye strain or nausea.
- Strong concerns that using it in real interfaces would be an “accessibility nightmare,” especially for people with low vision.
- Screen readers wouldn’t be helped by image-only text; HTML text with this font wouldn’t block AI anyway.
Use Cases and Motivations
- Ideas floated: memes to evade automated filters, anti-scraping, CAPTCHAs, anti-surveillance signage, or camera-/Ring-/Flock-adjacent obfuscation.
- Most commenters doubt its practical value and see it as easily bypassed.
- Some see it as part of a broader “AI resistance” impulse, more about pushback against abusive or exploitative AI use than against the tech itself.
Art, Novelty, and Critique
- Many praise it as a clever, visually interesting art/hobby project rather than a serious security tool.
- Critique centers on overstated claims about “stopping AI,” plagiarism accusations (contested), and general skepticism about turning such tricks into real products.