1 points fny | 1 comments | 22 Oct 24 14:24 UTC | HN request time: 0.205s
| source Sudoku is NP-complete. I've tried several puzzles generated by a tool[0], and ChatGPT manages somehow to solve many of the puzzles with the prompt "solve this sudoku puzzle, do not run code." It does fail from time to time, but its still remarkable that in O(n) steps it's able to get so close.
Similarly, ChatGPT doesn't struggle at all with sorting short sequences. With long sequences, however, only o1 produces valid results consistently.
I'm curious to hear peoples thoughts on what might be happening under the hood.
[0]: https://qqwing.com/generate.html