Google Research's Algorithms & Theory team recently outlined a perspective on incorporating mobility into language models. The central idea is that movement patterns can provide a richer, more grounded understanding of place than text alone. Instead of treating locations only as names or coordinates, models could learn from how people and objects move through spaces.
The post argues that mobility captures something about place that static text does not: how locations relate to one another through use, proximity, and human activity. For language models, this could mean better reasoning about spatial context and the everyday meaning of places.
Only one source was reviewed for this article, so there are no conflicting views to compare. The piece reads as an early-stage research argument, with the practical payoff still to be demonstrated.