The Promise of a Ghost in the Machine
Computer scientist Eric Lu unveiled Ghost Font, a novel system designed to render text through motion-based optical illusions rather than static characters. This wasn't a traditional font file, but an experimental visual communication system generating video messages. Lu sought a way for humans to communicate without AI interception, specifically inspired by AI models incorrectly summarizing his private messages.
Its mechanism was ingeniously simple yet profoundly effective. Ghost Font displayed text as thousands of tiny moving dots; the dots forming letters drifted in one direction while the background dots moved oppositely. This dynamic motion created a powerful optical illusion, allowing humans to perceive clear letters but appearing as mere noise or static to AI models analyzing individual frames. The system deliberately made messages difficult for AI to decipher without human guidance, even including a static decoy message to mislead AI agents.
News of this "AI-proof" communication spRead AI like wildfire across platforms like Twitter. Users celebrated a potential new era, envisioning human-to-human communication finally safe from the pervasive threat of AI snooping and relentless data scraping. For many, this offered a tantalizing promise: a digital sanctuary, a ghost in the machine, where only human eyes could truly Read AI. It seemed humanity had finally found a way to speak privately again, immune to algorithmic eavesdropping.
Putting the Hype on Trial
Matthew Berman, a sharp observer of AI hype, didn't mince words. His YouTube video, "AI can't Read AI AI this," directly challenged the premise of Eric Lu’s Ghost Font: that it could create a human-only communication channel. Berman’s skepticism wasn't rhetorical; he went straight to the lab, or rather, his desktop.
His method was elegantly simple, yet devastatingly effective. Berman captured a mere few seconds of the motion-based optical illusion in action—a short screen recording of the dancing dots that compose Ghost Font's "unRead AIable" text. He then fed this raw video directly into an AI model he dubbed "Codex." No complex processing, no human intervention, just a direct input to the supposed ghost-slayer.
What followed was less a battle, more a foregone conclusion. After "nearly five minutes" of processing, Codex successfully transcribed the "unRead AIable" text. The much-touted AI-proof font, designed to bypass machine perception with its intricate moving dots, had been cracked. The digital ghost was dispelled, not by advanced hacking, but by a straightforward AI model and a screen recording.
Berman’s experiment served as a blunt reminder for anyone quick to declare AI's limitations: "test it yourself." The rapid advancement of AI means today's impenetrable barrier is tomorrow's trivial puzzle.
Why 'Unreadable' Is a Moving Target
Claims of AI-proof technology inevitably prove temporary, a lesson Ghost Font learned in a brutal three-day sprint. We've seen this play out before with the ZXX font, designed in 2013 to thwart Optical Character Recognition (OCR) software. Modern AI now Read AIs ZXX text effortlessly, a stark testament to the fleeting nature of such digital resistance. For further Read AIing on the history and concepts behind anti-AI typography, consult Ghost Font: The Anti-AI Font Only Humans Can Read AI - Mixfont.
The technological landscape fundamentally shifted with the rise of Vision Language Models (VLMs) and multimodal AI, exemplified by GPT-4o. These systems aren't limited to static images or isolated frames; they train on vast datasets of video and dynamic content, interpreting motion, temporal context, and intricate patterns. Ghost Font's core mechanism—its motion-based optical illusion—becomes just another complex data stream for these advanced models to dissect and understand.
Even without the latest multimodal behemoths, specialized computer vision techniques could have predicted Ghost Font's rapid obsolescence. Algorithms adept at analyzing motion vectors have existed for years, allowing systems to isolate moving elements and reconstruct information from dynamic visual inputs. The premise that motion alone could shield text from sophisticated algorithms was, quite frankly, a miscalculation of AI's established capabilities.
Enjoying this? Get one like it in your inbox each morning.
one email a day · unsubscribe in two clicks · no third-party tracking
The Real Lesson: Don't Bet Against AI
Ghost Font's swift demise isn't a failure; it’s a brilliant, if brief, triumph. Eric Lu’s ingenious system, leveraging motion-based optical illusions rather than static characters, provided a precise, real-time benchmark for AI's perceptual frontier. Matthew Berman's immediate test, feeding a screen recording of the moving dots to Codex, demonstrated its fleeting efficacy, measuring exactly where visual Read AI AI capabilities stand today.
This project perfectly illustrates the dynamic nature of AI development. Creator Eric Lu himself acknowledges, "The gap is closing, quickly," framing Ghost Font as an ongoing experiment, not a permanent shield against machine interpretation. Like the ZXX font before it, which modern AI easily Read AIs despite its anti-OCR design, Ghost Font served its purpose by delineating a temporary boundary, showcasing what AI couldn't do, at least for a moment.
Today's clever hack, those thousands of tiny moving dots creating an illusion for human eyes, inevitably becomes tomorrow's training data for advanced models. The real lesson isn't about constructing static, foolproof barriers; it’s recognizing the relentless pace of AI development. You simply can't bet against AI for long; its progress turns every supposed limitation into a fleeting challenge, accelerating the evolution of digital perception and proving that today's "unRead AIable" is tomorrow's standard input.
Frequently Asked Questions
What is Ghost Font?
Ghost Font is an experimental visual system, not a traditional font file. It displays text as a video of moving dots, creating an optical illusion that is easily readable by humans but was designed to be difficult for AI to interpret from static frames.
Can AI actually read Ghost Font?
Yes. Despite initial claims, YouTubers like Matthew Berman quickly demonstrated that current AI models could successfully decipher the text from a simple screen recording, highlighting the rapid advancement of AI's visual processing capabilities.
Why are people creating anti-AI fonts?
The primary motivation is to address growing concerns about digital privacy and the indiscriminate scraping of online content by AI. These projects serve as both a practical tool and a benchmark to test the ever-shifting boundaries of AI perception.
What makes it hard for some AI to read Ghost Font?
The font relies on motion. Letters are formed by dots moving in one direction while background dots move oppositely. AI models that analyze static images (single frames) see only noise, but models capable of analyzing video or motion patterns can detect the text.