The last return of the viral verrals makes “reverse location” by photos

There is a little about the tendency trial tendency: People are using chartpt to understand the spot showed in pictures.

This week, Openi has released their new models ai, o3 and o4-miniBoth that can unify “Reason” through loaded pictures. In practice, models can be wedding, and zoom on on photos – even wins and distorted – for analysis well.

These analysis ability, coupled with the ability to search the web, make it for a place of the power of the power. X users quickly discovered that O3, especially, it is good enough to deduct Cities, Points of referenceand look restaurants and bar by a visual clue disturbed.

Wow, caught and even a tree in view. pic.twitter.com/bvcoe1fqt

– SWAX (@SWAX) April 17th of 2025

In many cases, models does not appear to be drawn on “Memories” past chatgt conversations, or Esif dataWhat is the metadata attached to photos that reveal details as where the photo has been taken.

X is full of examples of users giving charmpt Weather restaurant, Neighborhood quarter, Facesand it authorityand instructing O3 to imagine playing: “GeoguessR,” a game online that challenges the rappie by Google Street Views.

This is a fun chatter function of o3. GeoguessR! pic.twitter.com/hcmixs8yd

– Jason Strires (@VYrofe) April 17th of 2025

Is a potential privacy problem. There is nothing that prevents a bad actor from screenshotting, say, the instagram story of a person and using chatgpt to try to do.

o3 is insane
I asked a mine friend to give a random photo
They gave me a random photo that take into a library
o3 knows in 20 seconds and it’s right Pic.Twitter.com/0k8Dxifokoy

– Yumi (@incyuiu) isn’t 2 April 17th of 2025

Surely this could also do before the launch of the o3 and o4-mini. Techcrough RRANCE A photo number through O3 and a fair-free model without capcibility of the image, vittice my patterns-place of place. Surprisingly, Gpt-4o arrive at the same answer, correct as o3 more often than not – and took less time.

There was at least one instance during our short test when O3 has found a Gpt-4O place could not. It’s an image of a purple leader, an ink lithe bar, o3 he responded, it was a wardiamsbree passage, like gpt-4o has muted, a public pub.

It is not to suggest or3 is flicked in this sense. Many of our tests failed – O3 failed in a loop, unable to reach a response was confident, or volunteed a wrong place. Users on x notice, which o3 can be pretty away. in their deductions of the place.

But he huge illustrates some of the enving risks presented by more able There seems as few saves in place to prevent this type of “online” in chatgpt, and Opena, the company behind the chatgpt, has not addressed the problem in their Safety report for o3 and o4-mini.

We have reached Open for comment. We will update our piece if they respond.