Puzzle Challenge

23 views
Skip to first unread message

Eugene Freuder

unread,
Nov 4, 2023, 12:43:54 AM11/4/23
to Constraints
There is a World Sudoku Championship and a World Puzzle Championship:


I assume a CP Sudoku solver could easily beat the world Sudoku champion, but it might still be nice publicity to demonstrate this, and at the same time to demonstrate that this is something that a chatbot can have trouble with. 

I tested ChatGPT 3.5 on a "kids" 4 by 4 Sudoku and it got it wrong, not being consistent with the presented puzzle. Of course, when the mistake was pointed out it apologized and did get it right on the next try. :-)

More challenging would be a CP puzzle solver to tackle the more varied World Puzzle Championships, especially confronting a never seen before type of puzzle, and working with the same description given to human contestants. 

Here is the 9th World Puzzle Championship Qualifying Test from 2000:


The archive of the puzzles from the World Sudoku Championships and World Puzzle Championships is online (for a fee unfortunately, but perhaps the WPF would provide them to researchers). 


You might try to work with the World Puzzle Federation to enter a computer program in the 2024 Championships. Or you might try to work with your national organization:


I'm thinking of something analogous to Matt Ginsberg's Dr. Fill's win at the American Crossword Puzzle Tournament:


There is also an online "Grand Prix" in the Sudoku and in the general Puzzle categories. Perhaps a computer program could enter. 


And there are apparently now WPF ratings and rankings:


Perhaps the WPF would be interested in hosting a computer competition, with individual or national computer competitors. . 

Another possibility would be to have a computer contestant version of the Sudoku and/or Puzzle Championships at CP. Perhaps invite the human world champions to attend, "commentate", and/or compete with the computer winners. And invite the press. :-) 

Anyone intererested in pursuing any of this? Collaborating with others? Perhaps something to engage your students?

Eugene Freuder

unread,
Nov 4, 2023, 2:15:38 AM11/4/23
to Constraints

I'm afraid I got some of the links wrong in my prior post. I've corrected them below. 

Also, I realize there may be a "Goldilocks" issue here. Once a puzzle is modelled as a CSP,  can a human be at all competitive? But enabling a computer to extract appropriate models from descriptions meant for humans for an arbitrary set of competition puzzles seems like a lot to ask, especially if the competitions can introduce novel puzzles. But perhaps some "just right" challenges can be formulated for human/computer or computer/computer competition. And let's not forget the possibility of teams of collaborating humans and computers. :-) 

Also I suppose it is possible some of the general puzzles are not constraint satisfaction problems -- but then isn't everything a constraint satisfaction problem? :-)

Eugene Freuder

unread,
Nov 4, 2023, 2:52:55 AM11/4/23
to Constraints
And of course I realize relevant work has already been done, e.g. https://ojs.aaai.org/index.php/AAAI/article/view/27072
Reply all
Reply to author
Forward
0 new messages