New York Times games are hard: A computational perspective | Hacker News Reader

New York Times games are hard: A computational perspective

arxiv.org

73 pointsPaulHoule33 comments

I do Wordle, Pips, Strands, Connections & Sudoku. Of all of them, I find Connections definitely the hardest, even without the occasional US cultural references that I miss.

I still think the ultimate puzzle is the Sunday crossword (followed closely by Thur-Sat), though Connections is great. And definitely difficult (but never feels unfair).

I cancelled my subscription a few years back due to the way NYT was covering the current administration. At the time, I believed they'd never offer a "puzzles only" subscription because then they'd lose a large part of their subs. But, I was wrong. And now they offer a puzzle-only subscription.

There's a great documentary about the Crossword with Will Shortz that came out about a decade ago that's interesting.

Spelling bee is also pretty consistent.

In 2023, 55% of visits to NYT's website were to games, not news. The puzzle-only subscription points to the NYT's fate as a game company that also offers news, much as airlines are credit card/loyalty point companies that also offer flights.

So my hunches were correct (majority of people were subbing to play the games), but not my conclusion: that they wouldn't split it off.

I've always thought that Will Shortz was one of the most powerful people at the NYT (slightly joking, but sorta not).

I would say the Saturday puzzle is definitely harder. Sunday’s is just bigger.

100%. Every now and then they'll throw out a particularly tough Sunday, but I've yet to do a Saturday that wasn't difficult.

And if you didn't know this, Thursdays and Saturdays can have rebuses.

Just Connections, Wordle, and Mini for me (in that order), with the occasional Crossword (tend towards a barbell strategy of just doing maybe Mon, Tue, Sun to get the quick hits and a real challenging puzzle).

Also experience the odd difficulty due to Americanisms, but can't really fault a puzzle coming from something called the New York Times for that. I do however think the puzzle setting for Connections is inferior to The Wall from Only Connect, where they got the idea from. If you haven't seen that yet it's definitely worth a watch (it gets harder as as a season progresses).

I think the hardest part of connections is that there are intentionally overlaps between most (if not all) of the categories. When my wife and I would do them daily, she would steadfastly refuse to make a single guess until she was fairly confident in the entire solution because in practice it was hard to be sure of the current four for a given category even with high confidence of what the category is. (I agreed with her but in practice have both less patience and more trouble figuring out certain categories, so I would often guess and either take a few misses before figuring it out or sometimes completely fail).

Connections are the only one that can't be solved by either a regexp grep against /usr/share/dict/words or an LLM query. I was actually surprised how poorly LLMs fared against it---I thought they'd be better at associating peripheral connections.

There is a theory that for a puzzle game to interesting it has to be NP-hard. Something about how otherwise your brain is too good a latching onto the "trick" and the game is boring.

the rubiks cube is in P space, but has a large state space

towers of hanoi is also in P-space, has a trick to latch onto, and is still popular - though maybe this strays from being interesting and is popular for different reasons

> towers of hanoi is also in P-space, has a trick to latch onto, and is still popular

Is it? It's popular to introduce people to, and it's fun to play with for a bit, but once you understand how to solve it, there's basically no value in replaying it.

I see short form content vids occasionally of people doing large ones. i wouldnt say no value to replaying it. the value is more of a catharsis akin to power washing or cleaning things up i think than it is out of puzzle-like interest. so not a great example but i was looking for the simplest feasible example

Wordle, Connections, Spelling Bee, and Pips (All modes) for me. Wordle is fun but Pips is very satisfying. Pips medium can sometimes be more difficult than hard one.

I do wordle, strands and, when available (once a week) I do the midi.

Wordle and strands together usually take less than 5m. The midi ranges from 3m30 (my best time) to ~12m (my worst).

Not done Letterbox, Pips and Tiles, but I figured that all their puzzles are at the same level of difficulty.

It's interesting, to me, that (from my reading of the paper, which was very quick) they they consider it hard/easy based on a sort of brute-force attempt to find all the answers.

What's the midi? Or do you mean the Mini (Crossword)?

> What's the midi? Or do you mean the Mini (Crossword)?

There's 3x sizeof of crosswords

1. Large Standard New York Times one (subscriber only) 2. Mini (subscriber only) 3. Midi - which is in-between the large and the mini.

Once a week, the Midi is available to non-subcribers for free. A link comes in via email if you are signed up for their games.

Letter boxed is quite difficult if you want to do it in 2 words (which should always be possible), often taking me well over an hour. But they accept you 'winning' with 3-4 or sometimes 5 words. It's my favourite game as a non-native English speaker since knowledge of other languages is extremely helpful for optimal solutions. Especially as I speak French, Dutch, German and studied Latin and ancient Greek, don't think Swahili or Mandarin would help much.

If you enjoy the NYT games but want something new too, check out The Daily Baffle at https://dailybaffle.com. There's a range of word and logic puzzles that NYT lovers should appreciate.

If someone wants a challenging Wordle variant, I've made this: https://squareword.org

It's a combination of 2D Wordle, crosswords and Sudoku. Been running for over 4 years :)

Very nice!

I made a simpler take on "challenging Wordle" some years ago: https://quuxplusone.github.io/wordle-clone/evirdle/ (Try "easy mode" too!)

For clarification, these results mean the problems are difficult when some aspect of the problem size grows (e.g. dictionary size, alphabet size, ...). For example for letter boxed, the size of the square can vary, so can the alphabet size and dictionary of words. See Table 1.

It is not really meaningful to talk about the computational complexity of most problems exactly as they are published in NYT, or they end up trivially in P, since the problem description length is bounded by finite English letters, fixed board size, finite English dictionary etc.

They are difficult. As I'm not a native English speaker, I didn't know many of the obscure words or usages, so I actually played these games from a purely computational perspective. I discovered early on that there were a lot of at least NPC problems in them. As my English improved (partially thanks to these games), intuitions began to help me take shortcuts, as if I had become a nondeterministic Turing machine.

> They are difficult. As I'm not a native English speaker

Kudos to you. That would be insanely difficult. There is a lot of American-based pop culture, knowledge, and slang that makes it even difficult if you are a native, English American speaker.

The recent alcohol-themed Strands was brutal for me as an American. The first hint I received was utterly incomprehensible.

To avoid spoilers, the word in rot13 is just as meaningful to me: Znyorp

lol I'll have to check it out. Thanks.

Yeah, it can be pretty difficult at times. I'm quite proud of my 75% solve rate with Connections, which I'm slowly but surely improving (though the last week or so has been a bit of a regression).

I'm almost tempted to include that stat in my next CV as evidence of my grasp of the language :p

I always find it interesting to take a look at the Connections Bot, which gives the puzzles a difficulty rating based on how many people solve it or fail. It's not rare that I nail one rated 5/5 difficulty, just to completely fail the next day on a 1 or 2 out of 5. The gaps in general knowledge that you can have as a non native can be pretty funny at times! The groups relating to sports team names always get me.

> I'm almost tempted to include that stat in my next CV as evidence of my grasp of the language :p

lol It actually is that impressive that you should.

> which gives the puzzles a difficulty rating based on how many people solve it or fail

Wow I didn't know that how that worked.

And again, I am insanely impressed non-native speakers can get through those games because they're difficult even if you do know the language.

OT but recently came across this incredible video of a very engaging solve of a very beautiful Sudoku modification.

If you like puzzles this will brighten your day.

https://youtu.be/yKf9aUIxdb4

I do World, Strands, Connections, and occasionally pips and the spelling bee. I used to be a big fan of the minis but you have to be logged in to do them now :(

> consider four of them not previously studied: Letter Boxed, Pips, Strands and Tiles.

Statistically, approximately zero people play Letter Boxed and Tiles.

I do Spelling Bee, Wordle, Mini and both Connections every day, the rest as time allows. I enjoy them all.

I'm honestly a bit surprised that strands has managed to stick around as long as it has in the current form. The way "hints" work especially mystifies me; I don't really get why someone would want an entire word given to them more than maybe once total just from being able to guess completely unrelated words that are in the puzzle.

Reply on news.ycombinator.com