Concept · part of AI in the chain
Picture a university that stops relying on whatever chatbot a tech company is selling this year. It trains its own assistant on its own vetted library, rewards it for being right instead of agreeable, runs it on its own machines, and has scholars check what it produces.
That picture is the book’s hopeful case for AI. The technology is the same as the commercial chatbot. The difference is custody: the institution holds the training text, the reward and the configuration, and makes those choices in the open. Used this way the model extends what the institution already does. It gives students patient unpacking of hard sources from the institution’s own archive, tutors at scale, and drafts material that experts with real depth then verify. It speeds the experts up and does not replace them.
The book is blunt about the odds. Curating the text, training from scratch and running the system is the most expensive infrastructure the project has had to consider, and the money cannot come from ads without recreating the problem (see funding decoupling). Its verdict: technically realizable, economically marginal, and politically blocked by the forces that shaped social media. Almost no institution or commercial model meets the spec today.
Read the full argument
- Chapter 12: Infrastructure for Integration, section “LLMs as a capability extender, conditional on custody”
- Chapter 11: AI as a New Kind of Node, section “The political economy decides”
How it connects
- Builds on: Decompression on demand
- See also: Corpus capture, Funding decoupling from attention markets, Preservation vs. training (paired institutional functions), Versatile expertise, Work of generations (institutional timescale)
- Where the book discusses it: Chapter 11: AI as a New Kind of Node, Chapter 12: Infrastructure for Integration
← Deployment capture · AI in the chain (7 of 7)
Explore all the ideas · Read the book in order · Open the interactive map