Concept · part of AI in the chain
With a newspaper, you know an editor picked the stories. With a search engine, you know something ranked the links. Ask an AI assistant a question and you get a clean, confident paragraph, with every choice behind it out of sight.
The book counts those choices. One company decides what text the model learned from, what it was rewarded for saying, the hidden instructions it runs under, and who pays what for which version. Then the model writes the answer itself, where a search engine would point to someone else’s. The first two choices set the model’s built-in assumptions, and each is a place where the data, the reward or the setup can be tilted. A social platform owns two of these levers. A language model’s owner holds five or more at once.
For the book this makes AI the limit case of its worry about who tunes the filters and what pays them. The same machine could offer patient explanation on demand or the tightest grip on what people know that any medium has had. Ownership decides which. The book notes that open models spread some levers around, and that encyclopedias concentrated several long before.
Read the full argument
- Chapter 11: AI as a New Kind of Node, section “What an LLM is, structurally”
- Chapter 11: AI as a New Kind of Node, section “Where I land”
How it connects
- Leads to: Compressed form becoming authoritative, Decompression on demand
- See also: Corpus capture, Deployment capture, Inductive bias, Objective capture
- Where the book discusses it: Chapter 11: AI as a New Kind of Node, Chapter 12: Infrastructure for Integration
AI in the chain (1 of 7) · Decompression on demand →
Explore all the ideas · Read the book in order · Open the interactive map