ai-phi #40: Primavera De Filippi - Beyond the Promptable Latent Space

​When we speak of the "latent space" of a generative AI model, we usually mean the part we can reach through prompts: the linguistically addressable region that text-conditioned generation makes available.

​For Primavera De Filippi, this promptable latent space is often homogeneous and aesthetically conservative. What we call "AI slop" is not simply a failure of the model, but a consequence of the interface and validation systems through which the model is made usable.

​This ai-phi session asks what happens when we move beyond the prompt: through direct latent sampling, latent direction discovery, image-based references, and other interfaces that make the model navigable without routing every aesthetic decision through language.

​At stake is not only a new way of making images, but a political question of cognitive sovereignty: if the promptable latent space defines what can be reached through AI, then who defines its perimeter?

​https://ai-phi.github.io/posts/session-40-beyond-promptable-latent-space/