Adversary · Paper 03 · 12 min
Generate the obvious — then kill it
The only honest use of a model at idea-time is to map the kill zone quickly, in public, so you are not tempted to live there.
You do not beat the model by ignoring it. You beat it by making it show you the field, then standing somewhere it did not suggest.
Thesis
Why this paper exists
Refusal without information is vibes. Refusal after a map is a hack.
- Run a kill-zone pass in ten minutes
- Keep only ideas that survive a mean constraint
- Stand one step off the map, on purpose
The ten-minute pass
Prompt: 'List the twelve products 80 teams will ship for this brief. No extras. No poetry.' Then: 'Now list why each one loses on Sunday afternoon.' You now have a map and a stack of obituaries. This is the highest-leverage ten minutes of the weekend.
Anything you still want to build must survive those obituaries, plus one constraint the model did not emphasise — usually a person at a minute.
Standing off the map
Off the map is not 'wilder'. Wilder is how you get a punchline without a product. Off the map is a specific embarrassment the brochure skipped. The dining hall is not short of bins. It is short of a way for a proud student to claim leftover food without looking poor.
If your off-map idea needs a new platform, you walked too far. Come back until it fits in a sentence a cook would say.
Do this in front of the intern
Show the model your surviving sentence and ask it to attack. If it can kill the sentence with 'that's just X', either accept the kill or change the sentence until that line does not fit. This is sparring. It is what the Spar room is for.
You are training a reflex: the model proposes, you dispose. Never the reverse.
Drill
Run the twelve-product pass on a brief you care about. Write obituaries. Keep one survivor that names a person and a minute. Ask the model to kill it. Rewrite until the kill does not fit.
Knowledge check