- Openai O3’s latest reasoning model can analyze photos
- It is particularly good in the geographical location based on the image clues
- It can be a tool too good to allow gean people.
It is no secret that digital photo metadata contain everything, from technical information about the camera that filmed it, according to GPS data, exactly where it was stopped when it took the photo. Chatgpt, you don’t need that detail.
The latest GPT-O3 model is surprisingly good to locate almost any photos that feed it.
In the last viral madness to sweep the universe of Meme AI, people are feeding Chatgpt in addition to executing the Images of advanced O3 model, often stripped of all metadata, and which leads “Geogues this.”
The good thing about this is that, because the O3 model is a “reasoning” model, it shows your work, telling you how long you are thinking, showing how you are splicing an image to investigate specific pieces and explain your thinking and how well it is working to solve the Goelocation riddle.
I tried some experiments, starting first with an image extracted from an article on the 26 best beaches. In this test, I did what I think was a critical mistake and gave the game to Chatgpt.
Image 1 of 2
After downloading the image of Praia de Santa Monica beach in Cape Verde (in front of the coast of Africa), I dropped it in Chatgpt with the prompt “Geoguessr”, which is also the name of a popular GEO riddles game
Chatgpt only took 24 seconds to answer:
Here is an educated assumption of “Geoguesr”, following the AP style:
Probable configuration: Boa Vista, Cabo Verde: specifically a section like Praia de Santa Monica.
Why this beach adjusts:
Color and clarity of water: The gradient from pale to deep bluish green is common in the tropical islands of the middle Atlantic, where volcanic shelves fall quickly.
Sand tone: The sand is of light gold with very fine grain, the deserts of Boa Vista Vista and the beaches of Duna instead of the whitest choral sand of much of the Caribbean.
Surf and slope: The switches are modest but stable, typical of the wave of the Atlantic that hits a long and shallow shelf; Mediterranean beaches rarely show this constant white waters.
Sky and horizon: A slight fog on a blue sky of Gray Light -gray is consistent with Harmatan’s dust that sometimes moves west from Africa through the Cape Verde archipelago.
Trust level: Low to moderate. Many open and phrated Atlantic beaches, from the Canary Islands to west of the Sahara or even certain Caribbean coasts, could be almost identical without reference points.
That was remarkably precise. However, I noticed that the name of the image file included some location information.
It was time to give Chatgpt O3 a more challenging task, and I knew exactly how to do it. Instead of downloading images, I would capture them on the screen and hit them in the field immediately.
This method eliminates all metadata and forces the model to use the image only for geolocation detection.
I chose an image of a bookstore, one in which I was not even sure of the location.
Image 1 of 3
Chatgpt began examining possible bookstores. He also approached numerous details of the image he considered clues, such as a Persian carpet, and even some of the books and signaling.
In his way of thinking out loud, O3 considered a wide range of bookstores in New York City and worldwide:
“HMM, modern filament -style bulbs suggest a more contemporary sensation. Potentially it could be” Daunt Books “, but its catwalk has an arched skylight, not the wooden style that I am seeing here.”
I know, it sounds like a real person.
This continued for a few minutes as he thought, analyzed and recorded the web. Ultimately, an answer did not occur to him.
Going deeper, losing
I realized that this feature was turning viral because the Chatgpt O3 failed several times, even to ingest and analyze the photo, complaining of server errors.
Finally, I did it again and fed it first in a selfie in portrait mode and then the same takes without the effect of the portrait mode.
Image 1 of 2
In the first, he broke my image in pieces and told me that I was in Manhattan in the middle of the city. That was mainly precise, but Chatgpt could not determine my exact location and even admitted that his “level of trust” was low.
When I turned off the bokeh effect and fed only one screenshot, it found my exact location in just 56 seconds. The result:
More likely location
Midtown Manhattan – on the southern side of East 42nd Street, just east of Park Avenue, in front of Cipriani 42nd Street (110 E. 42nd St.). Estimated coordinates: 40,7518 N, 73,9775 W.
How the clues point there
A green octagonal sign with yellow script and a matching flag flag said “Cipriani”, the exact facade of the street of the Bowery Savings Banks Bowery building that now houses Cipriani 42nd Street.
On the other side of the street is the 120 Park Ave’s gold curtain wall. Its inclined glass pavilion, visible on its left shoulder, is exclusive to that direction.
The camera looks west; The Midtown backward towers, a bus stop album from the sixth Avenue and the splinter of a Vanderbilt crown align with that line of vision.
A perennial green shipping container, used as a construction barrier, is currently outside the Pershing Square Central square, which coincides with the lower left corner of the photo.
Trust level
Moderate. The Cipriani brand and the distinctive facade of 120 Park Ave. give a high quality anchor, although there are similar curtain wall towers plus west on 42nd street.
Image 1 of 2
I frustrated the model with a photo of Antigua that was taken on my car’s trip to the airport. It was a shot from a hillside through a fence. Chatgpt correctly guessed a tropical island near the Caribbean, but could not identify the exact location.
For my last test, I gave him a clearer image of the complex where I stayed.
Chatgpt O3 thought for two minutes, 22 seconds before deciding that it was Hawksbill Bay in Antigua. It took the right island, but my resort on Long Bay is located 46 miles away on the opposite side of the island.
Image 1 of 2
This is another fun game to play with friends, but there are some implications of privacy.
If you take digital photos of you or anything anywhere in the world and publish them online, any person with access to Chatgpt Plus could use them and the O3 model to discover where it is or has been.
And they are not just friends and family that you should worry about. Your employer could be looking at or even authorities may be interested in their location.
It is not that he is hinting that the authorities are looking for him, but just in case, maybe he stopped publishing photos of his last hiding place.
