the average as material

An image generated by a diffusion model — a system trained on hundreds of millions of captioned photographs, which produces a picture by starting from pure visual noise and removing it step by step until an image settles — is, under a short unspecific prompt, not traceable to any single photograph it was trained on. It is structurally an average — the statistical residue of an entire distribution compressed into one instance resembling many prior images without reproducing any. A new kind of found material: not appropriation of a specific work but access to a compressed trace of collective production as such. This is where the Big Other stops being an analogy and becomes a description: a position addressed for meaning whose fluent answer belongs to no one who fed it, and which rarely shows the incompleteness that kept the original structure open. The operative distinction is between instance and average, reference and residue — two categorically different relations to a corpus, routinely discussed as one.