/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.