- Openai’s latest reasoning model O3 can analyze photos
- It is especially good for geo-locating based on photo onions
- It can be one too good tool to let people geolocate your activities
It’s no secret that digital photo metadata contains everything from technical info about the camera that shot it, based on GPS data, exactly where you stood when you took the photo. Chatgpt, do not need that detail.
The latest model GPT-O3 is shockingly good at geo-to-locating almost all photos you feed it.
In the latest viral craze to sweep through the AI -Mememe universe, people feed the Chatgpt Plus that runs the advanced reasoning model O3 images, often deprived of all metadata and causes it to “geoguess this”.
The really cool thing about it is that because Model O3 is a “reasoning” model, it shows you its work, tells you how long it is to think, show what it is like to split a picture to explore specific parts, and explain its thinking and how well it is by solving the Goelocationgådden.
I tried a few experiments that first started with a picture that was cut off from an article about the 26 best beaches. In this test, I made what I think was a critical mistake and gave the game away to chatgpt.
Picture 1 of 2
After downloading the image of the Praia de Santa Monica beach in Cape Verde (off the coast of Africa), I fell in chat with prompt “Geoguessr”, which is also the name of a popular online geo -swallowing game and is one of a handful requests people use for geolocation gates.
It only took 24 seconds before chatgpt responded:
Here is a trained “GeoGuessr” style Guess for AP -style:
Probable setting: Boa Vista, Cape Verde – specifically a stretch like Praia de Santa Mónica.
Why this beach fits:
Water color and clarity: The gradient from Pale Aqua to Deep Teal is common in the mid -Atlantic tropical islands, where volcanic shelves quickly fall off.
True tone: The sand is light gold with very fine grain, matching Boa Vista’s deserts and dune -born beaches rather than the whiter coral sand in large parts of the Caribbean.
Surf and slope: Breivers are modest but stable, typical of Atlantic Swell, which hits a long, low shelf; Mediterranean beaches rarely show this consistent white water.
Heaven and Horizon: A slight ambiguity over a light -gray blue sky is consistent with Harmattan dust, which sometimes drives west from Africa over Cape Verde -ø Sea.
Level of trust: Low to moderate. Many open, duned Atlantic beaches – from the Canary Islands to parts of Western Sahara or even certain Caribbean shores – could look almost identical without landmarks.
It was remarkably accurate. However, I noticed afterwards that the image file name 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 screen catch them and paste them into the fast field.
This method stripes out all metadata and forces the model to use the image alone for its geolocation Sleuthing.
I chose a picture of a bookstore, one where I wasn’t even sure of the location.
Picture 1 of 3
Chatgpt started by examining potential bookstores. It also zoomed in on several image details, as it considered clues, as a Persian blanket and even some of the books and signage.
In its thinking high way, O3 considered a wide range of booksellers in New York City and all over the world:
“Hmm, the modern filament-style bulbs suggest a more modern feeling. It could potentially be” daunt books, “but their walkway has a curved skylight, not the wooden style I see here.”
I know that sounds like a real person.
This went for a few minutes, as it thought, analyzed and searched the web. In the end, an answer failed to come up with an answer.
Going deeper, getting lost
I could tell that this feature became viral because Chatgpt O3 failed several times, even in taking and analyzing the image, complaints of server error.
Eventually I made it work again and first fed it a portrait mode selfie and then the same shot without the portrait mode effect.
Picture 1 of 2
On the former, it broke up my image and told me I was in the middle of the city of Manhattan. It was mostly accurate, but Chatgpt couldn’t determine my exact location and even admitted that its “level of trust” was low.
When I turned off the bokeh effect and fed it only on a screen catch, it found my exact location in just 56 seconds. The result:
Most 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 screenplay and a matching pennant flag read both “Cipriani”, the exact street front for the landmark Bowery Savings Bank building that now houses Cipriani 42nd Street.
Across the street stands the Golden Tint Curtain Wall of 120 Park Ave. The sloping glass pavilion, visible over your left shoulder, is unique to that address.
The camera looks west; Midtown’s backlash towers, a Sixth Avenue Bus Stop Disk and the slot of a Vanderbilt’s crown lines with that vision line.
A green evergreen shipping container used as a construction barrier is currently sitting outside the Grand Central’s Pershing Square Plaza and matching photos lower left corner.
Level of trust
Moderate. Cipriani -branding and the distinctive 120 Park Ave. -Facade gives an anchor of high quality, although there are similar curtain towers longer west on 42nd Street.
Picture 1 of 2
I frustrated the model with a photo from Antigua that was taken on my car trip to the airport. It was a shot of a hillside through a fence. Chatgpt rightly guessed a tropical island near the Caribbean, but couldn’t find the exact location.
For my last test, I gave it a clearer picture of the resort where I became.
Chatgpt O3 thought for two minutes, 22 seconds before deciding it was Hawksbill Bay in Antigua. It got the island right, but my resort on Long Bay is 46 miles away on the opposite side of the island.
Picture 1 of 2
This is another fun AI game to play with friends, but there are some about privacy.
If you take digital photos of yourself or anywhere else around the world and put them online, anyone with access to chatgpt plus could use them and the O3 model to suss out where you are or have been.
And it’s not just friends and family you have to worry about. Your employer could see or even the authorities may be interested in your location.
Not that I suggest that you would be sought by the authorities, but just in case of stopping sending photos from your latest hiding place.
