Recent developments in the field of AI have led to a large number of newly released tools based on large language models, such as ChatGPT, Bing or Bard. While these models are quite impressive, it has become increasingly clear that they don’t know everything. This raises questions about AI’s knowledge boundaries, the nature of latent spaces, and their accessibility.

knowledge spaces makes it possible to explore the embedded knowledge of AI models through an interface that provides the spectator with a platform to explore and search high-dimensional vector embeddings, visualised on a two-dimensional surface. Each high-dimensional vector embedding will be represented as a point in the visualisation, and each search will generate a new point on the surface, drawing connections between similar pieces of data in the embedding space. The resulting network of interconnected information will be visually similar to a map of an underground railway network, following recent studies that suggest knowledge is best represented as rich, interconnected networks rather than linear trees.

2023