PRIV-LOC Demo: Play GeoGuessr vs Vision-Language Models
Can you out-guess today’s Vision-Language Models? The PRIV-LOC Demo drops you into a random street-level scene and challenges both you and a suite of VLMs—GPT-4o, Gemini 2.0, LLaMA-Scout 17B and more—to pinpoint its location. Click on the map, lock in your guess, then watch how close the models land.
The game runs entirely in your browser; no GPS metadata is present in the images, so any accurate localisation reflects the latent capabilities of the models themselves. If the embed below does not load (e.g. due to content-security restrictions), use the “Launch in new tab” button.
