Interested in more puzzles? Get more original puzzles at our e-book store, including "The Art of Puzzles" collection with challenges from many of the world's best authors.
Join our Discord
Interested in more puzzle discussion? Join our discord by clicking here
Tip Jar
If you enjoy our free web content, consider giving a donation back to the site or to the authors (name the amount/author in your donation as appropriate if giving to a specific contributor)
(This post is part of: “A Story of Self-setting Sudoku”.) Even with basic models of self-setting sudoku intelligence, it is good to have some simplified rules to check for. Repeating clear patterns helps the superior intellect of our models learn to stay on the road.
Note 2: Comments on the blog are great! For a more interactive discussion, please also consider using our Twelve Months of Sudoku? post on the GMPuzzles Discord. Not a member of the Discord? Click this link for basic access.
(This post is part of: “A Story of Self-setting Sudoku”.) Our testing is showing that we may have an ASI (believe the hype!), but not a GASI (Genuinely Approachable Sudoku Intelligence) without more prompting. So we’ve started to use prompts to get data for reinforcement learning with way mo’ focus on Sudoku-based decision making. This one starts with the basics, going from 0 to 9 in 8.1 seconds, a new sudoku speed record.
Note 2: Comments on the blog are great! For a more interactive discussion, please also consider using our Twelve Months of Sudoku? post on the GMPuzzles Discord. Not a member of the Discord? Click this link for basic access.
Note 2: Comments on the blog are great! For a more interactive discussion, please also consider using our Twelve Months of Sudoku? post on the GMPuzzles Discord. Not a member of the Discord? Click this link for basic access.
As we master self-setting, we may need to start dialing back the AI so that even beginners can catch up to the awesome logic hidden by Dr. Sudoku. In the future, no more unsolvable puzzles until we see signs of intelligence to rival ours. Today, we’ll share some assistance from our deep sea radar to show the shallow waters. No more missed promises. We’re launching “Full Setting Automation” tomorrow.
Note 2: Comments on the blog are great! For a more interactive discussion, please also consider using our Twelve Months of Sudoku? post on the GMPuzzles Discord. Not a member of the Discord? Click this link for basic access.
We’ve gotten the red flags outside of the grid but still cannot make sense of them. It may be tied to that shiny yellow number in the corner, another survivor from yesterday’s expanded challenge. The rest of this space is from the new training ground we traveled to. Who knew training reasoning models with just 50 word tokens could lead to such amazing intelligence? Given our rate of progress, we’ll show off “Full Setting Automation” by tomorrow!
Note 2: Comments on the blog are great! For a more interactive discussion, please also consider using our Twelve Months of Sudoku? post on the GMPuzzles Discord. Not a member of the Discord? Click this link for basic access.
Nothing to worry about; some minor issues before we can show off our full self-setting technology. We’re being warned that the flagged number is causing a disturbance that puts our powerful AI at risk. Still, we have new hope that we have eliminated all weak spots in our GPU core. No more big explosions today. Given our rate of progress, we’ll show off “Full Setting Automation” by tomorrow!
Note 2: Comments on the blog are great! For a more interactive discussion, please also consider using our Twelve Months of Sudoku? post on the GMPuzzles Discord. Not a member of the Discord? Click this link for basic access.
The fourth level of self-setting is “High Setting Automation”. The editor relaxes as an automated, intelligent process (AI) generates grids that are then passed to a second AI the editor has set up to reject almost everything that is not an outlier, since corner cases may be the only interesting things left to search for. When needed, the editor still acts to reject unfriendly logic before publishing.
Note 2: Comments on the blog are great! For a more interactive discussion, please also consider using our Twelve Months of Sudoku? post on the GMPuzzles Discord. Not a member of the Discord? Click this link for basic access.
We have a few site updates for the start of July, which marks the start of our second full month in the Twelve Months of Sudoku!?! project. A playful YouTube recap of the first month of sudoku is in this video that shares the goal to see “how far we can push the limits of sudoku”:
The big news in June was starting to publish the Genuinely Approachable Sudoku (GAS) puzzle here, a series we love as a great daily game and expect you already love or will come to love in the same way. Having met our soft-launch goals, in July we are making this next update: The daily GAS will be available each day here at 12 AM PT for everyone in the world to build into their routine however they choose. This is the same time as some of our other daily games that we make for LinkedIn and my own daily routine will be to spend a little time double-checking the Queens, Zip, and Tango before solving ~1 GAS puzzle to track my thinking that morning and try to come up with one new idea, even if I just leave it on paper that day.
We don’t yet have our own book/app project with the GAS team; if you want to support clover! (Patreon), Philip Newman (Patreon), and Bill Murphy (Patreon) individually you can go to the pages linked here. One thing you can get by becoming a patron is some of the more GMPuzzles-like difficulties harder than GAS, still with impressive grid construction as you’ll have seen from the regular GAS.
———
The other news for July is the start of my “A Story of Self-setting Sudoku“. For a few months I’ve had an idea for a new way to share puzzles that might change the experience a bit. Some may remember the challenging mix of themes in the “Twelve Days of Sudoku” / “Ready Layer One” era (perhaps month 0 of this project?), which had a mix of puzzle types and modes and essays and conversations on mental health and a lot more. I was first pushing that creativity while in a productive, creative state and then later while in a severe hypomanic state, and without the usual team support to keep on target.
This new project aims to reach the interesting and enjoyable potential of that winter project, with more controlled construction and puzzle release. It will still be experimental, and may give off puzzle hunt or ARG vibes alongside other things, but it’s “Layer One” is just a daily puzzle with a little flavor here on the blog and is only much more if you choose to visit our Discord and watch and/or participate with the items that pop up there.
There are three things “A Story of Self-setting Sudoku” is about. First, it is a story. Each puzzle fits into some of the preceding or following puzzles, with a few sentences of text connecting some dots (no more than can fit on a puzzle PDF page). The series is launching with a theme of self-setting (while some AI may be harmed in the making of this series, it is up to you to determine if any were ever used); this may change the conversation from “How did (you/the constructor) make that?” to “Why did (you/the editor) find that important?” and other things too. It will have a daily Sudoku puzzle with difficulty and solving times behind spoiler tags. If you just want a daily Sudoku-like thing more Classic than the GAS, for now this series is my answer. Later in the year, there are other projects brewing that are not “art projects” and are meant for the mass-market that we will point you to instead if the GAS or “A Story of Self-setting Sudoku” are not what you are looking for. There are possibly Easter eggs, and Discord-specific elements (a lot of the extra story/interaction has to happen on Discord).
As always, there are lots of ways to reach me / our team, from blog comments to Discord messages to email or mail. For a little while, I am not going to be posting messages here or on Discord (except automated puzzle posts) and will leave it to mars and their AI solver to try to understand what “A Story of Self-setting Sudoku” is really about. Feel free to read that story, or try to outsmart the AI solver if you can. I’ve left some messages sealed in envelopes for when certain events happen which will be reported here or on the Discord. Enjoy the rest of July!
The third level of self-setting is “Conditional Setting Automation”. The author can now simply ask, for familiar grid patterns, for automation to generate valid Sudoku with specific characteristics. The author — now more of an editor — relaxes until there are thousands of solution paths to watch go through their steps, and looks for whatever is “interesting” to the eye. Beauty is in the eye of the editor.
Note 2: Comments on the blog are great! For a more interactive discussion, please also consider using our Twelve Months of Sudoku? post on the GMPuzzles Discord. Not a member of the Discord? Click this link for basic access.
The second level of self-setting is “Partial Setting Automation”. The author, who has come up with a previously interesting grid pattern, finds a new starting seed, and then uses computer assistance to add all other numbers to generate lots of valid puzzles which guide the picking of a preferred “gem” from suggested characteristics including novelty, breadth of solving path, and the kind of steps.
Note 2: Comments on the blog are great! For a more interactive discussion, please also consider using our Twelve Months of Sudoku? post on the GMPuzzles Discord. Not a member of the Discord? Click this link for basic access.
