- Author
- René Walter
- Published
- 14 May 2026
- Kind
- Long personal newsletter essay, four parts collected in one post
- Original
- goodinternet.substack.com/p/on-interpolatable-archives-clean
- Subject
- Art and images
What it is
René Walter writes the Good Internet newsletter on Substack. This is the clean version of an essay first published in four parts. It sets out what Walter means by an interpolatable archive. It then lists uses and risks.
What it says
- Walter says language models are interpolatable archives. They blend their contents.
- The essay reads Aby Warburg's picture atlas as a prototype of an image model's space.
- Walter argues models are cultural technologies, not agents. Seeing agency in them is a fallacy.
- It reads a chatbot solving a 60-year math problem as retrieval of a dormant answer.
- Walter names homogenization, a flattening of collective creativity, as possibly the most serious pitfall.
- It rejects the parrot and bullshit critiques as arguments against meaning in models.
- It ends ambivalent on labor, energy and democracy, and urges regulation over rejection.
What evidence it gives
The essay offers argument and many citations to papers and writers. Walter also gives a personal account of using a chatbot to research consciousness theories. There are no experiments of the author's own.
Interests and limits
This is one writer's opinion piece, and the cited studies were not checked here. Walter says the clean version removes the slop and some silly parts. The author uses chatbots in personal work. The essay admits it leaves labor, energy, water and democracy questions open.