Posting a 5x5 crossword every day | Hacker News Reader

Posting a 5x5 crossword every day

crucig.com

47 pointsldom2230 comments

Keyboard navigation needs some improvement. It's not easy to navigate because you are blocked from correct guesses. Also, it wanted me to guess "SASHA", but the hint was for "Sacha" (Baron Cohen).

Not just navigation, but typing too. Just typing a word results in letters overwriting themselves because the next letter comes in before it's transitioned to the next space, so you have to type artificially slowly.

"American figure skater, last name Cohen" would have worked as the clue.

Most of these entries are Naticks where the two crosses are proper nouns. That’s really unpleasant as it usually means there’s no way to figure out the cross unless you know the two terms.

https://en.wikipedia.org/wiki/Rex_Parker

In 2008, he invented on his blog the crossword term "natick" (after Natick, Massachusetts) for an "unguessable" square crossed in both directions by proper nouns considered obscure.

I love the NYT mini so I made some scripts and HTML to auto generate more 5x5 puzzles, posting one very day at my site

Did you auto generate the clues?

Kind of. They are auto gen from Wikipedia content but I edited them manually so they were shorter and made more sense

Sounds like an llm task.

Great effort!

I'm a fan of the NYT Mini, too. The art of that puzzle, I think, is in the clues, which are usually not straight synonyms or dictionary-type definitions. Some examples from recent puzzles:

  Name tag heading: HELLO

  Word shouted during a defibrillator scene in a hospital drama: CLEAR

  Kind of orange with a "belly button": NAVEL

  "I'm not a ____" (online affirmation next to a checkbox): ROBOT

  What breweries might creatively repurpose as seats: KEGS

I haven’t tried [1], but it might be hard to get an LLM to produce such clues consistently. I’m not sure how well I would do at it myself.

Another feature of the NYT Mini is that the creators seem to limit the number of pop-culture facts to at most one per puzzle. I’m bad at names of actors, cartoon characters, pop stars, etc., but I’m usually able to solve the puzzles even when they have clues and answers like

  Ogre who asks "What are you doing in my swamp?!": SHREK

because I can get the answers in the other direction.

[1] A few minutes later: I did try. I gave Claude 3.5 Sonnet the above clues and answers as examples and asked it to produce clues for today’s NYT Mini answers. Here are the results:

https://gally.net/temp/20240904miniclues/index.html

Not as clever as the NYT’s clues, but not terrible, either.

Could you share some details on what algorithm you use to generate the puzzles?

Crosswords are one of those interesting things where there is a quite a bit of trans-atlantic difference which is worth taking a brief look at if you aren't familar. To quote Wikipedia:

> Crossword grids such as those appearing in most North American newspapers and magazines consist mainly of solid regions of uninterrupted white squares, separated more sparsely by shaded squares. Every letter is "checked" (i.e. is part of both an "across" word and a "down" word) and usually each answer must contain at least three letters. In such puzzles shaded squares are typically limited to about one-sixth of the total. Crossword grids elsewhere, ... have a lattice-like structure, with a higher percentage of shaded squares (around 25%), leaving about half the letters in an answer unchecked.

I grew up with Swedish crosswords, where the clues are written into the "non-letter" cells (the ones that are shaded in an American crossword). There's no separate clue section, it's just all part of the grid. They almost always have some illustration as well, they look really delightful compared to American crosswords, though from a "puzzle" angle they're not always as fun.

I think that might be called an 'Arrow word' in at least some English dialects, if I am getting your description correct in my head...

E.g. see: https://images.app.goo.gl/mErh6pEEnR9Eexpf8

Yes, that kind of thing, but with an illustration and usually styled nicer. This is a very typical look (the first crossword i found by googling): https://www.argument.se/wp-content/uploads/2018/10/Korsord-2...

In Spain they're called Autodefinidos: https://es.wikipedia.org/wiki/Autodefinido

this is one of my frustrations as a brit, i much prefer the US format, but find that US puzzles contain so many 'Americanisms' that they're much harder than they might otherwise be.

Still do the NYT every day though.

I do the NYT mini most days too. I consider solving the Americanisms an additional part of the challenge !

This is great! Thank you for sharing.

I have been running a 5x5 game that is more wordle like (i.e. no clues, guess entire words) at https://squareword.org for the past ~2.5 years. Interestingly, when you exclude proper nouns, there is a relatively limited amount of possible 5x5 squares with common words - I am having to add some uncommon words to make it work. I guess with nouns included this is less likely to happen, should be a much larger space of possible words.

I play your game almost daily for a year now. Keep up the good work.

Thank you, it makes me happy to hear that!

Neat idea (editing UX need a bit of work, perhaps); I would much like to see _all_ hints in parallel on my screen.

Solving a crosswords is a bit like solving an equation system, since the grid is equivalent to all row questions AND all column questions combined in a big mega-confunction, so it helps to see all at once.

That's nice ! Learned a few things today, it's more effective than hitting "random article" on wikipedia. It would be nice if you could add an icon to the site, so I could find it more easily in my bookmark toolbar.

You can just type randomly and finish it up. It displaying correct letters and locking them when you enter one makes it a bit easy to cheat

Fantastic! This should be a 'Show HN'

Which part of Belfast are you measuring from? Google maps says its only about 55 miles...

Can I see the clues after I have solved it?

This is too hard to be fun.

The fact that it tells you if individual letters are right or wrong makes it much easier.

Exactly. I even started making an auto-solver based on that. I stopped because I have more important things to do today and I don’t even enjoy crossword puzzles (no shade on anyone, you do you).

Love this!

Reply on news.ycombinator.com