ChatGPT will get scarily good at guessing photograph areas, sparking doxxing issues
A scorching potato: Now that individuals have principally stopped utilizing ChatGPT to show themselves into motion figures, it appears the following development involving the AI is utilizing it to guess areas based mostly on pictures. Whereas some are discovering this reverse location search performance enjoyable, it raises a number of privateness issues, particularly in the case of doxxing.
OpenAI launched its newest o3 and o4-mini fashions final week, which may “motive” by way of uploaded photographs. This implies it will probably crop, rotate, and zoom in on pictures, even when they’re of poor high quality.
Mixed with the fashions’ different skills, individuals have discovered that they’re significantly good at figuring out areas in uploaded pictures.
Customers are feeding o3 photographs of every little thing from restaurant menus to selfies and telling the mannequin to think about it’s taking part in the web guessing sport GeoGuessr, which duties gamers with guessing areas based mostly on Google Avenue View photographs.
Wow, nailed it and never even a tree in sight. pic.twitter.com/bVcoe1fQ0Z
– swax (@swax) April 17, 2025
It is easy to see this as all enjoyable and video games, however there is a probably darker facet. This reverse picture search might simply permit somebody to be doxxed – the general public revealing of the place they reside or are positioned – based mostly on minute particulars in a picture that almost all people wouldn’t discover. A easy selfie with few background objects, or a narrative on social media, may very well be fed into ChatGPT to study the place it was taken.
Whereas customers have praised the o3 mannequin’s capability to determine areas from photographs, it is not one thing that arrived with the newest releases. TechCrunch notes that GPT-4o, which was launched with out picture reasoning, was capable of give you the identical solutions as o3 as a rule, and it did so in much less time. Nevertheless, there was one occasion within the publication’s testing the place o3 was capable of appropriately guess {that a} image of a purple rhino head mounted in a bar was from a Williamsburg speakeasy – GPT-4o thought it was from a UK pub.
It is essential to notice that even o3 does not get its guesses proper each time, and typically it will get caught in a loop when attempting to find out a location.
An OpenAI spokesperson mentioned that visible reasoning will make its instruments extra useful in areas like accessibility, analysis, or figuring out areas in emergency response.
As for stopping doxxing, the spokesperson mentioned the fashions refuse requests for personal or delicate info, and the corporate has added safeguards supposed to ban the fashions from figuring out non-public people in photographs.
Masthead: Alex Shuper
