/ˈleɪtənt speɪs ˌæktɪˈveɪʃən/The excavation of hidden vectors.
Definition
The process of prompting an AI model to access low-probability, high-creativity regions of its potential
output. Without specific "activation" techniques (constraints, style blending, perspective shifts), models
regress to the mean, producing generic content. Activation forces the model off its worn paths into the
"wild" regions of its latent space.
The Art of the Specific
To activate latent space is to refuse the default. It is the act of guiding the model away from the
cliché center and towards the strange periphery.
Field Notes & Ephemera
Field Note: If you ask for a story about a cat, you get a generic cat. If you ask for a
story about a diabetic cat who runs a pawn shop in 1920s Berlin, you get literature.