In a new viral trend, users are discovering ChatGPT’s ability to identify locations from photos through its latest AI models, o3 and o4-mini, which feature advanced image reasoning capabilities. These models can analyze images by cropping, rotating, and zooming, even on blurry or distorted photos. This ability, combined with the models’ web search function, allows ChatGPT to deduce locations like cities, landmarks, and even specific businesses such as restaurants and bars. The trend quickly gained traction on social media, with users uploading photos and challenging the AI to identify locations, mimicking the online game GeoGuessr.
While these new AI capabilities offer intriguing possibilities, they also raise significant privacy concerns. Users have shown how easily ChatGPT can determine the location of places like restaurants or even private settings, like personal photos, without using metadata like EXIF data. This could lead to risks such as doxxing if a malicious actor were to upload an image from social media and attempt to pinpoint someone’s whereabouts.
Though prior AI models, including GPT-4o, could also identify locations, the new o3 model is notably faster and more accurate in many cases. During testing, o3 correctly identified the location of a bar based on subtle details in a photo, whereas the older model made a less accurate guess. However, o3 is not flawless—some tests showed it could not determine a location accurately, and other users noted that the AI sometimes provided incorrect answers.
The trend underscores the potential dangers of more capable AI models, as OpenAI has not yet fully addressed the privacy implications of these new features. While the company has included safeguards to prevent abuse and protect privacy, the situation highlights the need for further caution in the development of AI technologies. OpenAI has responded by emphasizing its commitment to privacy and monitoring misuse.
