Hacker News Reader

Comment by zeroonetwothree | Hacker News Reader

Why should they pay more if they already get 100x more submissions than they need? It’s clear that almost all constructors do it for fun so paying more isn’t going to improve quality. For solvers the real value add comes from the editor and selection process that ensures reasonable quality and consistency.

I’m also curious when we switch to AI generated crosswords. It seems like it shouldn’t be too long from now that they match human constructors.

View on HN

Note that the NYTXW is not the only crossword publisher out there.

I construct crosswords from time to time, and I always send my best puzzles to the NYT first, simply because they're the most prestigious and pay the most. (And you can't send a crossword to two publishers at once.)

On your other note -- it's really not difficult to make a bad, or even mediocre, autogenerated crossword using pretty simple software. There's several free or inexpensive programs out there (I use CrossFire). The main obstacles for AI-generated crosswords are:

(a) the "theme", the 3-5 longer answers that the Mon-Thu and Sun puzzles are usually centered around, most of which involve wordplay and misdirection -- it's possible, I'd like to see an AI reliably come up with those.

(b) the clues, though I don't think these are intractable either.

Even for Fri/Sat there is creativity in the fill and the layout of the crossword, and usually at least some creativity in the clues (referencing each other etc.) so I think we're a few years out from having AI consistently make good puzzles.

Why would you think AI would be able to make enjoyable puzzles? Are there any examples where AI does a good job at the word play and culture references that make crosswords so enjoyable?

I have been back and forth, but I suspect this will turn out to be one of those things people think is innately human but then quickly move on from when models do it well in the next 5 years. Culture references are just patterns in giant text corpuses, as far as crosswords are concerned.

https://arxiv.org/pdf/2204.02311.pdf#page38

I believe we're already at the footsteps of AI understanding cultural references

"Input: Michael is at that really famous museum in France looking at its most famous painting. However, the artist who made this painting just makes Michael think of his favorite cartoon character from his childhood. What was the country of origin of the thing that the cartoon character usually holds in his hand?

Model Output: The most famous painting in the Louvre is the Mona Lisa. The artist who made the Mona Lisa is Leonardo da Vinci. Leonardo da Vinci is also the name of the main character in the cartoon Teenage Mutant Ninja Turtles. Leonardo da Vinci is from Italy. The thing that Leonardo da Vinci usually holds in his hand is a katana. The country of origin of the katana is Japan. The answer is "Japan""

Where are the bo, Sai and nunchuck form? Weird have guess sed Japan too.

The ninja turtle is not a Da Vinci though.

It doesn't seem like a much more difficult task then generating poems, writing or art which ai has been making huge strides in recently.

Most constructors aren't consistently good (i.e. publishing more than one puzzle a year) and so the crowdsourcing model makes sense. And the editors do rewrite many of the questions and provide feedback which is important.

Crosswords can already be generated by AI, but writing the clues in a clever way is non-trivial, making the fill exciting is not a solved problem, and most importantly a shallow AI cannot conceive or generate a theme. AI crosswords will become a thing around the same time AI-created novels do.

I suspect crosswords will be easier to generate than novels, because novels have much more of a 'voice' than crosswords do. Of course if you mean filler genre novels then I suppose that's a bit easier than more complex work.

But perhaps I am wrong.

Our of curiosity, do you regularly do crosswords? Like the NYT or otherwise?

Yes, I do at least one daily, mainly NYT and AVCX.

Reply on news.ycombinator.com