Puller investigates the legibility of AI vector space by mapping model-generated associations onto publicly available online sources. Its research boards place latent relations in contact with citations, making the passage from machine knowledge to documentary evidence available for inspection.

Demonstration board with fictional entities and sources.