A first-person demo narration where the speaker tests OpenAI’s (ChatGPT o3) image geolocation capability. After removing embedded GPS by screenshotting and attempting metadata extraction with Python, the speaker ran searches and cross-references; within ~43 seconds the system identified the exact spot — the Shirakawa Canal — provided coordinates and matching reference images. The clip emphasizes the tool’s surprising accuracy and privacy implications.
Chapters
OpenAI just gave everybody the ability to track down any photo on earth. I'm not exaggerating. Check out this example. So I looked at my camera roll and found this little snap of this beautiful little bird in a river here. And I just figured, how could it know this? I mean, it's just a river with some rocks and a generic looking Japanese building in the back. Well, long story short, it got this right, but I thought it was really interesting that initially pulled up Python and wanted to extract the metadata from the image, which I thought of already. I screenshotted the original image to make sure that there's no GPS coordinates embedded to make this actually challenging. So that didn't end up working. Made some assumptions, ran some searches, looked at all the websites, cross referenced this, and after only 43 seconds, it concluded that this elegant gray heron is standing in the Shirakawa Canal. And then it gave the exact coordinates and a description of where you can find this exact spots with other reference images that look similar to this. And yep, that's the exact location where I took this picture. God damn. How is it this good?
