alt Hacker NewsYeah, I came up with the original idea entirely independently back in 2023 (https://gwern.net/idea#zen-sand-garden-puzzle-game) and postponed it until LLMs could code it up for me (ie. now), and Claude/ChatGPT turned that up as prior art while researching it: https://gwern.net/doc/design/2026-10-03-gwern-samon.html#mec...
ZPG winds up not being as interesting, I think, because it doesn't figure out how to take the core idea of raking and twist it into more interesting puzzles, which is the classic puzzle approach: find a clean core mechanic, and then escalate on it.
FWIW, I wish you'd spent a few minutes cleaning up the first dozen pages of Claudespeak. You have some human text at the bottom of the post, but first you force your readers to wade through (or skim, or just close the tab) stuff like "Solver experiments show the bare rule is loose" — "is a chore with the shape of a puzzle:" — "test candidate mechanics on thousands of random gardens with an exact solver, a constraint-programming optimizer, and a deductive difficulty grader" — "the only U-turn-free raking among 12 par rakings of 6,310,226. Ripple stones and wave bands, the obvious “pattern” constraints, add texture but not tightness" — ...
You do see how that's all garbage, right? So why not just delete it? Keep the good part, delete the bad part.
The game widget itself takes up 3 pages on mobile, because of all the Claude web design.
... Wait, as I scroll down again in composing this complaint, I realize I'm wrong about there being any human text at all. It's all garbage. This makes me kind of sad for the Gwern we've lost.