o3 model goes viral for spotting real-world locations
OpenAI revealed two new reasoning models for its ChatGPT tool earlier this week, claiming they can “think with images.” The o3 and o4-mini models can interpret and manipulate images, fetching information to improve output.
The models have sparked a viral trend among expert users, who have discovered they can geolocate photos with limited additional inputs. The o3 model, in particular, appears proficient at identifying locations in images, even with obstructions.
It can edit images by cropping or zooming to extract information, responding with precise geographical coordinates and the name of the place. This ability works with both outdoor and indoor images.
Experts, including Wharton associate professor Ethan Mollick, have confirmed that the model doesn’t simply pull geotagged information from photos but instead uses its own reasoning. However, like any AI model, it’s prone to incorrect responses, especially with limited cues.
The model’s geolocation ability has raised concerns about potential misuse, as it makes the process effortless for anyone with access. Without barriers to prevent unauthorized use, determining someone’s location could become a significant privacy risk.
Some examples of the model’s capabilities have been shared on social media, with users demonstrating its ability to identify locations. As X user Jason Barnes shared, “nice. saying it was in az might have been too big of a hint.
While geolocating has been a niche activity among experts, ChatGPT’s latest update makes it accessible to a broader audience, raising concerns about responsible use.