Hacker News Reader

The next step is to make crossword puzzles from this.

(I'm next to illiterate about constraint satisfaction programming. How hard is to make reasonable crossword puzzles?)

The constraint programming part isn't the bit that makes the difference between a good and bad crossword. The difference is in the quality of the clues. Generating plausible 'cryptic' clues is probably well beyond the ability of current AI.

If you wanted 'Jeopardy'-style clues, that's easier.

Creating a grid to fit the chosen words in wouldn't be so hard. The real difficulty would be coming up with clever clues that weren't just the first paragraph of the wiki article with the names removed

Reply on news.ycombinator.com