Large Spatial Models

Language models struggle with geometry and physical structure. The ask is large spatial reasoning models that treat 3D structure as a first-class primitive.

Large language models drove most recent AI breakthroughs, but mainly in domains that can be expressed through language. Today's systems handle simple spatial tasks such as depth estimation, but they can't reliably reason about 2D and 3D features, spatial manipulation or mental rotation, which limits AI in the physical world.

The ask: large models built around spatial reasoning, where geometry and physical structure are core building blocks rather than something approximated through language, so AI can reason about and design real objects and environments.

Written by Ryan McLinko of Astranis, a YC company.

Read the full request on YC →