“I would have written a shorter letter, but I did not have the time.”
Most writing used to pre-expand ideas for every possible reader or was imperfectly verbose. Given recent AI advances, we can publish a compact core and let AI expand it per reader on demand.
Compress verbose text, audience-conditioned decompression.
Human nature and audience uncertainty has historically created verbosity in final write ups. In this new era, humans can express their idea with minimal energy. AI compresses drafts to a core while preserving intent. The same core decompresses differently for background, purpose, and context.
Mathematically: traditional writing optimizes $\max_y E_X[\text{Comprehension}(X; y)]$, one expression $y$ for all readers $X$. AI enables $E_X[\max_y \text{Comprehension}(X; y)]$, optimal expansion per reader. By Jensen’s inequality,
$$E_X[\max_y \text{Comprehension}(X; y)] \geq \max_y E_X[\text{Comprehension}(X; y)].$$Longform remains relevant for style, persuasion, and art.
Humans as novelty injectors
Humans supply the new bits. Fresh observations, opinions, and creative leaps. AI filters redundancy against the historical record, keeping only the novel core and reattaching shared context during expansion.