‹ Culture 3.14 · What we ask ourselves
Question · Language
Which part of a narrative should be generative and which deterministic?
The same piece contains data that admits no variation and explanations that need it.
Why it matters
Generating the whole piece with a model makes every figure depend on the drafting. Writing it entirely with templates produces content nobody quite finishes reading.
The boundary between the two is not aesthetic: it decides what can be checked automatically and what needs review.
What we know so far
Figures, dates, names, units and explicit causal relations do better deterministically, inserted from the structure. Transitions, narrative order and adaptation to the reader’s level work better generated.
When the model writes over a typed structure instead of over an earlier draft, correcting one piece of data propagates on its own to every version.
What remains open
The boundary blurs in comparison and in implicit causality, which is where the value of a report is concentrated.
And we do not know how to hold that separation when the same piece has to exist in text, voice and video with different durations.