A New Trend That Raises Eyebrows

A concerning trend is sweeping across social media: users are leveraging ChatGPT to uncover the locations depicted in photos. This phenomenon has gained traction since OpenAI introduced its latest AI models, o3 and o4-mini, which can analyze uploaded images in remarkable detail.

Unpacking Image Analysis Capabilities

These cutting-edge models boast the ability to crop, rotate, and zoom in on even the most distorted images. This powerful functionality, combined with their web searching capabilities, creates a formidable tool for pinpointing locations. Enthusiasts on X (formerly Twitter) are discovering that o3 is particularly adept at deducing everything from cities and landmarks to specific restaurants and bars, all from subtle visual hints.

Privacy Concerns on the Horizon

The implications of this trend are chilling. With just a simple screenshot from a platform like Instagram, a malicious actor could potentially use ChatGPT to track someone’s whereabouts. This form of ‘reverse geolocation’ raises significant privacy issues, especially since it allows for more invasive discovery tactics.

Comparing Performance: o3 vs. GPT-4o

TechCrunch conducted an intriguing comparison between o3 and the previous model GPT-4o, which lacks image reasoning capabilities. Surprisingly, while GPT-4o often matched o3's accuracy in determining locations, it did so more swiftly. In a specific test, o3 identified a dimly-lit bar featuring a purple mounted rhino head as a Williamsburg speakeasy, while GPT-4o mistakenly identified it as a pub in the U.K.

Not Without Flaws

However, o3 is not infallible; in several instances, it failed to provide reliable answers, getting stuck in loops of uncertainty or offering incorrect locations. Users on X have reported similar inaccuracies, indicating that while the model is advanced, it's far from perfect.

Emerging Risks of AI Models

This new trend highlights the emerging risks accompanying advanced AI systems like o3. Alarmingly, there seem to be minimal safeguards in place to prevent this reverse location search from becoming widespread. OpenAI has not explicitly addressed these concerns in their safety report for the new models.

We've reached out to OpenAI for their perspective on this unsettling trend, and we will update this article if they respond.