Show HN: ChatGPT sees something else in this image (gavin.wtf)

A developer showcased a novel method to manipulate how AI models like ChatGPT interpret images, revealing that AI can perceive entirely different content than humans. Concerned about the misuse of personal likenesses by AI, the developer experimented with animated PNGs to conceal information. The result is a technique where the visible frame presents one image while a hidden frame contains a contrasting interpretation that ChatGPT accesses when provided the original file. This innovative approach holds significance as it underscores the evolving capabilities and vulnerabilities of AI models in image processing. By leveraging features of animated PNGs, the developer demonstrated a way to obscure original content from AI, although limitations exist, such as file size constraints and the potential for browsers to alter the images during copy-paste actions. While this method doesn’t fully address concerns of likeness misuse, it opens up discussions about image security in the AI/ML community and highlights the ongoing challenges of ensuring ethical AI usage.