OpenAI’s newest fashions, launched this week, can decide the placement of photographs utilizing context clues, which OpenAI calls a “vital breakthrough in visible notion.”
Early testers are importing photographs and asking ChatGPT to “geoguess” the place they have been taken. The outcomes of this “reverse location search” are surprisingly correct.
The AI analyzes the photograph and spends just a few seconds “pondering,” or analyzing the context clues inside it. It then gives a solution together with “a protracted inside chain of thought” about why it got here to that conclusion. Within the instance under, a person took a photograph of a library e-book, and it accurately guessed that it was taken on the College of Melbourne based mostly on a code on the label.
This Tweet is currently unavailable. It might be loading or has been removed.
One other person uploaded a generic photograph of a house in Suriname, which seems to be like it might have been pulled from Google Earth, and ChatGPT will get it proper. It is just like the AI model of this guy.
AI-powered photograph locators have been round for a while, however the model utilizing OpenAI’s o3 mannequin appears to have popularized it for the plenty. That mentioned, the earlier mannequin and present flagship, GPT-4o, has the identical functionality, albeit with decrease accuracy.
We examined it by asking each fashions to guess the placement of a picture we took this week at the New York Auto Show of Subaru’s latest EV, which debuted at the show.
GPT-4o wasn’t capable of pinpoint a selected location, however it accurately mentioned that it is perhaps at an auto present, in both Chicago, New York, or Los Angeles. It based mostly that on “the setting: polished show atmosphere, a number of autos in shut proximity, informational signage, and folks strolling round observing the automobiles.” It incorrectly learn the title on the automobile and referred to as it the “Trailspeed,” not “Trailseeker.
GPT-4o guesses that the photograph was taken at an auto present, however would not know which one, and will get the title of the automotive fallacious. (Credit score: ChatGPT)
Get Our Finest Tales!
Your Each day Dose of Our High Tech Information
By clicking Signal Me Up, you affirm you’re 16+ and comply with our Terms of Use and Privacy Policy.
Thanks for signing up!
Your subscription has been confirmed. Keep watch over your inbox!
The newer o3 mannequin obtained it proper. It “thought for 1m 40s” after which defined that the “blue crossover is Subaru’s new 2026 Trailseeker EV, a mannequin that was first revealed to the general public on the present stand on the 2025 New York Worldwide Auto Present (NYIAS) inside Manhattan’s Jacob Okay. Javits Conference Middle.”
It crawled Subaru’s automobile launch web page to substantiate the Trailseeker debuted on the present, and paired photos of Subaru’s sales space design with the photograph, discovering a match with the “lighting, carpeted ‘forest‑flooring’ motif.”
OpenAI’s o3 mannequin accurately guesses the place the photograph could have been taken. (Credit score: ChatGPT)
ChatGPT may mix picture recognition with picture manipulation. If customers add an imperfect picture, ChatGPT can transfer parts round to reply questions on it. Within the instance under, it deciphers considerably illegible writing written the other way up on a pocket book.
Advisable by Our Editors
“I am going to must load the picture so I can examine the textual content. As soon as I view it, I notice the textual content is the other way up, so I’ll rotate it so it is readable,” ChatGPT explains as a part of its ‘thought’ course of. “From there, I can test what’s written and share my findings clearly with the person.” It determines the writing says, “4th February – end roadmap.”
(Credit score: OpenAI)
OpenAI says the mannequin “can nonetheless make primary notion errors” and that “even when tool calls accurately advance the reasoning course of, visible misinterpretations could result in incorrect remaining solutions.”
Different apps additionally use AI to establish the placement of photographs. Geospy, for instance, makes use of context clues equivalent to vegetation and structure to find out location. It made headlines earlier this yr when 404 Media reported that it might be exploited by regulation enforcement and stalkers since customers may ask ChatGPT to geolocate a photographs posted on social media.
OpenAI imagines the tech might be “useful in areas like accessibility, analysis, or figuring out places in emergency response,” an organization spokesperson tells us. “We have labored to coach our fashions to refuse requests for personal or delicate data, added safeguards meant to ban the mannequin from figuring out non-public people in photos, and actively monitor for and take motion towards abuse of our utilization insurance policies on privateness.”
About Emily Forlini
Senior Reporter
