We wanted to share a “the story so far” kind of update for “A Story of Self-setting Sudoku” since a lot of people have not been following very deeply. Importantly, if you just like solving really elegant Sudoku puzzles, please just solve the Sudoku. You can ignore the author, title, and flavor text; the provided difficulties suggest what you’ll be getting on any post. However, the posts do link together so…

For the light summary (web view only):
In the first week, Thomas shared a set of 4 (of 5) puzzles on a box-like grid, titled “Intelligence is Thinking * The Box” which each had very different solving structures. The flavor text suggested some may have been hand-set but by the end he was possibly enjoying time at the beach and a computer was doing most of the work. While getting to full self-setting AI was problematic, adding in new training sources besides just Classic Sudoku got some more curious puzzles and better AI. Finally, on Monday, the “Dr. Sudoku” system launched on its own, exceeding the typical 9×9 space of a sudoku puzzle.

We’ve been told that each week (roughly) there will be a new theme for the Sudoku, and this is now the second day for an apparent “self-driving” theme. The puzzles seem easier so far — at least today’s only needs 1-6 placed — but who knows what is down the road.

For the heavy summary (web and discord rabbithole):
Besides the story above, the president of GMPuzzles, mars, has been using their own HAL 12000 AI system to analyze and solve the puzzles. They have been finding occasional traces of puzzle hunt-like signals in the sudoku, and with the help of Tane, HAL was able to hack into the Dr. Sudoku system but has since gone quiet.

We await further word, but this might be a battle of intelligence against intelligence. We’ll need help eventually. There is a spoiler-free place in the Discord just for the HAL 12000 messages and a cooperative discussion space. But first you may have to join.