Concept · part of AI in the chain

You ask an AI model to summarize a research paper in three bullet points. The bullets arrive in seconds and sound sure of themselves. You do not open the paper, and neither does anyone you pass the bullets to.

The book’s claim is that this is new: no earlier compression carried this kind of authority. A magazine piece presents itself as one popularization among many. An AI summary presents itself as the answer, with the machine’s authority behind it. At scale the summary stands in for the source in what people collectively remember: the paper still exists and nobody reads it. It is the old warning about maps and territories, with a map so convenient that the territory goes unvisited.

Two things drive it. The summary costs nothing to get while the paper costs hours, the same price gap that starves depth elsewhere. And the model writes without the hedging a human expert has, because it never felt where a field’s knowledge runs out. So a compressed version, which may keep or distort the truth (the book describes three outcomes), arrives with no sign of what was dropped. It is the mirror image of decompression on demand.

Read the full argument

How it connects


← Decompression on demand · AI in the chain (3 of 7) · Corpus capture →

Explore all the ideas · Read the book in order · Open the interactive map