‘Pokemon Go’ players revealed to have been training game developer’s new AI model
By Bryan Ke
“Pokemon Go” developer Niantic announced its new AI model on Nov. 12, created using data players have sent over the years.
The large geospatial model (LGM), built using the San Francisco-based studio’s visual positioning system (VPS), helps computers “perceive, comprehend and navigate the physical world in a way that will seem equally advanced.” LGM is reportedly similar to large language models (LLM) like ChatGPT. Instead of text, however, Niantic’s LGM collects geolocated images from “Pokemon Go” players exploring locations to capture Pokemon.
The data-collecting feature, which can be disabled, is indicated in Niantic’s privacy policy, where the studio confirms it uses geospatial technology and player recordings to “build a 3D understanding of real-world places, with the goal of offering new types of AR experiences to our users.”
So far, the company has received over 10 million scanned locations globally using its VPS from the past five years and around 1 million fresh scans every week, “each containing hundreds of discrete images.”
Share this Article
Share this Article
