What it does
His best setup solved the easy puzzle; medium and hard ran out of rounds. The study used three puzzles, one run each, with thresholds chosen on those same puzzles. Code still checks contradictions and the final answer.
Benchmarks & research
jev-sudoku
Mark Santolucito tests whether Jev can rule out Sudoku candidates, comparing multiple-choice questions with separate yes-or-no questions.
Bookmark: jev-sudoku
Keep this in your collection.
Leave a noteWhat would you try with this?
: jev-sudoku
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
How you can use it
Use the reported test as a design example: compare asking AI to pick one digit with asking if each digit could fit. A developer can build a separate test for your puzzles. Live Jev answers need a TypeSafe access key. Check new puzzles before trusting the settings used in this small study.