The AI revolution started in earnest when the massively parallel power of GPUs let us train larger, deeper models than we could have dreamed previously. Since then, the models have been growing in size and sophistication. Now the new bottleneck is data. ChatGPT, image generation, TikTok's recommendation engine all demonstrate the incredible power of the new algorithms when the data is there. But spatial tasks are generally missing that crucial data.

Let's explore how AR apps, useful in their own right, can be leveraged as an unprecedented source for spatial data to feed a new generation of spatially intelligent models. We will touch on both performance and ethical concerns associated with passive data collection, as well as best practices for pipeline and deployment.