Tomtit in the tilth -- the NY TIMES SPELLING BEE thread

I don't even know if Semantle is about meaning, or letter clusters, or parts of speech or what.

it's semantic similarity? Quoth the maker again:

I thought about adding some graphs, but the words are in a 300-dimensional space and that is hard to graph.
I am not sure how to describe what the percentages represent. Like, if I say cosine similarity, that doesn't help. The answer is something like semantic similarity (with some grammatic similarity), and I can't really think of a better way to play than by thinking of semantically similar (or different) words.

i have no idea what that means ¯\_(ツ)_/¯

lol to be frank, neither do i! he's kind of brilliant and i can't always keep up with him.

abanana, what's not working for you on the share button? it should copy results to your clipboard.

I don't even know if Semantle is about meaning, or letter clusters, or parts of speech or what.

― Tracer Hand, Wednesday, February 2, 2022 8:38 AM (one hour ago)

The algorithm/model is aiming more for meaning than anything else, but since it cannot itself actually understand meaning, it will seem weird to a human is about how I would attempt to explain it. AIUI you train a model like this on a massive corpus of text and it "learns" both which words tend to be associated with others (like how auto-complete works in texting) and which words seem substitutable for others. Like if the model sees "I drive the car" thousands of times and "I drive the automobile" thousands of times, it "learns" that car and auto are semantically close. The model also "learns" that drive and car are often associated with each other. What I don't know is if/how it distinguishes between the synonym relationship of car/auto and the grammatical one of drive/car

Wow I got it in 78. Feel faintly lucky.

dmac, I think that Trayce gets it for me— in a sense, Wordle is more about about numbers and strategy, and that isn't my forte— words are, tho.

If you give me a pool of letters, I'm happy to make words out of them. Give me a clue for a word, and I am pretty good at figuring it out.

But if you give me five blank spots and say "guess a word that can fit in these spots," it doesn't keep my interest for more than a few minutes because there are so many five letter words that I have few fucks to give about whether I can get one or not.

I am good at Scrabble and similar because I can approach the essential numbers-based logic of the game through a pool of letters at my disposal. Not having anything at my disposal, there are no limits, so it seems pointless.

each guess removes letters from your pool though. the limits increase with each turn, pointing you closer to the solution.

FUCKS
XXXX?

Yeah it's very rare that you aren't steered closely enough to deduce the word. it sort of reminds me of those codeword puzzles that are essentially blank crosswords but with a number substituted for each letter. Or hangman, as someone said above.

W/e, I just don't like it. Make it have its own thread.

I thought I was going to hit the jackpot with COCKADOODLEDOO today, but alas.

Lol. Can't quite get to genius today, yet. Taking a break.

Oh wait. Only two points away. So close and yet so far. Halfway to paradise.

One point away. But EELCAKE is not accepted for some reason.

Okay made it with the second pangram

Is Semantle supposed to tell you the answer if you press “give up”? It didn’t for me.

Last time I played it didn't even show my guesses.

got semantle in 28 today!

it should tell you the answer after giving up.

Will pass along these notes. i am mystified at how you guys are getting better at this. I can't do it at all!

Todays SB does not allow TRYHARD as a word and I am annoyed.

tantara was the word from today that I hadn't been aware of before

re: semantle

at least on a laptop, when you click 'give up' a dialogue box appears asking you to confirm

I haven't gotten better at it and haven't figured out how to get closer to the answer, I'm choosing words that are like other words on the top of my list, but haven't gotten above 50%, and choosing random words doesn't seem to work

I think the tricky part is that the words at the 50% mark really aren't that much like the answer, so you have to keep casting your net in new directions even though it looks like you're getting closer.

I guess, but there are infinite directions

I got up to god and buddha today, my two choices with the highest similarity (only 21-27%), but the answer was candle, which is only remotely related to those words, and I wouldn't have guessed it

semantle creator response

The answer is printed just under the guess box -- did they not notice it? Maybe I need to give it a border or something.

Re "for 200", I did break it for about 8 hours one night earlier this week (oops), but it has been working since except for three or four 15-second outages.

There were color bars today, for the first time? Yes?

It's very confusing that you can be at oh idk like 50% but also scoring 986 out of 1000. I don't understand the metrics.

I gave up on yest and tdays but only because it's late and I want to do other things. I was going down the wrong path with todays--like Lily said, the answers around 50% were not that close to the answer...I assumed they were leading me in a direction but they were more like a spoke of the same wheel and I should have kept turning it.

it's interesting to me that there seems to be a sort of perception capable of understanding this but I simply don't have access to it at the moment.

The 'getting close' bar on Semantle is a welcome addition. It can still be a frustrating game, though. Once I have guessed the word (or given up) I find it interesting to keep guessing at words that I know are related to the secret word. Some that you think should be extremely close are lower down than you might imagine.

What was yesterday's semantle? I got to popular which was 997/1000.

(was going to carry on trying but forgot)

Ah, I can't recall. In recent days all I can remember are elevator, boil, impatience, overnight, and eager. 'Overnight' was ridiculous because, of my guesses, 'day' scored higher than 'night', so I never would have gotten it if I'd kept trying.

My god can y'all start a new thread for Wordle, Semantle, etc.

Yes, was wondering.

sheeeesh, fine
Semantle: Can you think like a computer?

thanking you, forks. sorry for being brash.

can’t believe JUMANJI and MANGINA are not accepted

Ha!

That J words the first thing I tried lol

bye bye Wordle stats. boo.

Seems not everyone lost theirs, but I guess I'm glad I wasn't the only one.

I find it mildly interesting to see my guess distribution so it would be a bit of a bummer to lose those but apparently ppl are annoyed that their winning streaks are erased & I’m not sure it’s worth being bothered about those with a game this easy? Like it feels virtually impossible to lose with 6 guesses

Yeah the guess distribution was the one I found most interesting.

they’re getting rid of guess distribution??

looks like nyt officially took over yesterday, and I saw ppl on twitter posting nyt formatted stats (including guess distribution + streak intact) so i think it may just be a glitch due to the crossover?

im really close to my modal guess becoming 3 so i really hope they don’t wipe the distribution

xp - no, they didn't get rid of it, it just seems like the stats for a lot of people were reset with the switchover

https://www.devangthakkar.com/wordle_archive/

day 3 was a rough start

boringly obvious pangram today