Harvest Kernel Interactive Kit
The Core Grammar Drill
80 minutes to a number almost nobody using AI can produce: your personal miss rate.
A randomized trial in JAMA Network Open handed fifty physicians GPT-4 and their diagnostic scores moved two points. The model alone beat them by sixteen. The gap was judgment, and judgment is built from foundations. This drill trains exactly that layer against your own AI output. Work through the four phases in one sitting or across a week. Your entries live on this page only until you leave, so use the export button when you finish.
Overall progress 0%
Pull one claim
Take one thing AI handed you this week that you accepted and used. A number, a citation, a summary, a recommendation.
Find the probability under the certainty
Ask the model how confident it should be in the claim and what evidence would change its answer. Watch the verdict dissolve back into a forecast.
Build the case against
Argue the claim is wrong, using the model as a sparring partner rather than an oracle. If you cannot construct one decent objection, you did not understand the claim well enough to accept it.
Verify and log the verdict
Chase the claim to a primary source. Then log it. Ten entries in, the miss rate below becomes the most honest measure of your AI use anywhere.
| # | Verdict |
|---|
Your miss rate so far: no entries yet
Post your miss rate in the community
Members are comparing logs in the SeedStacking Resources space. Bring your number, especially if it embarrassed you.
Companion to AI Raised the Price of Not Knowing Things. Framework adapted from Manu Kapur’s core grammar argument in Times Higher Education. Plant Ideas. Cultivate Skills. Harvest Results.