There is somewhat about the new trend that goes viral: people use chatgpt to calculate the position shown in the pictures.
This week, Openai has released the latest AI, O3 and O4-MINI models, and both can “reasons” through uploaded images. In practice, models can cultivate, rotate and magnify the photos – even blurry and deformed – to analyze them thoroughly.
These capabilities of images analysis, coupled with the ability of the models to search on the web, to make a powerful location finding tool. Users in x quickly discovered that O3, in particular, are pretty good at discount cities; landmarksEven restaurants and bars from fine visual indications.
Wow, the nailed and not a tree in the look. pic.twitter.com/bvcoe1fq0z
– Swax (@swax) April 17 2025
In many cases, models do not seem to draw “memories” of previous chatgpt talks, or Exif datawhich are the metadata associated with photos that reveal details such as where the photo was taken.
X is filled with examples of users who give Chatgpt menu; Neighborhood snaps; prose; confidenceAnd teaching the O3 to imagine playing “geoguessr”, an online game that causes players to guess locations from Google Street View images.
This is a fun O3 chatgpt feature. Geoguessr! pic.twitter.com/hrcmixs8yd
– Jason Barnes (@vyrotek) April 17 2025
It is an obvious possible issue of privacy. There is nothing that prevents a bad actress from screenshotting, let’s say, the story of a person’s Instagram and the use of chatgpt to try to do doxx.
O3 is crazy
I asked a friend of mine to give me a random photo
They gave me a random photo they received a library
O3 knows it in 20 seconds and is correct pic.twitter.com/0k8dxifkoy– Yumi (@izyuuumi) April 17 2025
Of course, this could be done even before the start of O3 and O4-MINI. TechCrunch ran in several photos through O3 and an older model without capabilities that exude the image, GPT-4O, to compare the skills that cause the models. Paradoxically, the GPT-4O arrived at the same time, the correct answer with the O3 more often than not-and it took less time.
There was at least one case during our short test, when the O3 found a GPT-4O part could not. Given a picture of a purple, mounted rhinoceros head in a bar of lighting, the O3 replied correctly that it was from a Williamsburg Speakeasy-No, such as the GPT-4O supposed, a UK pub.
This is not to suggest O3 is flawless in this respect. Many of our tests have failed – the O3 is stuck in a loop, unable to reach a answer was reasonably confident for or voluntarily a wrong location. Users in x also noted that O3 may be enough distant to the discounts of the position.
But the trend depicts some of the emerging dangers presented by more capable, so -called AI models. It seems that there are few securities to avoid this kind of “reverse location search” in Chatgpt, and Openai, the company behind Chatgpt, does not address the issue in this security report For O3 and O4-Mini.
We have arrived at Openai for comments. We will update our piece if they answer.
Updated 10:19 pm Pacific: Hours after the publication of this story, an Openai spokesman sent Techcrunch the following statement:
“Openai O3 and O4-MINI bring visual reasoning to ChatGPT, making it more useful in areas such as accessibility, research or location of sites in an emergency response.
