Semantle: Can you think like a computer?

Definitely done that, yes! Or played while half-watching TV and typed words from the TV in.

Once I'm quite close, it takes me anywhere between a few guesses to several hundred guesses. On average? No idea but probably tens. The mean is probably 100+ after a couple of outliers (including the one flagged with a warning) where I didn't get it for ages after hitting the 990s.

this game is fun!

219 for me. Did that thing of being completely stuck and out of guesses despite being on 977. Went back to it later and it only took a few more.

I got up to 977/1000 and got fucked off + gave up and damn it I should have spent at least another half an hour, because was so nearly there. But it's all good experience for getting a feel for how this word2vec scores the word similarity. If I ever see anyone claiming today's wordle was hard I'm going to link this to them from now on, it's child's play in comparison.

I managed it in 31 today. Seems like low similarity scores (<20) are words that might appear in the same sentence as the target word (how big of a content window is the algorithm using?), medium scores (21-40) are words that commonly appear adjacent to the target word, and medium/high scores (>40) start trending towards actual synonyms or members of the same semantic category?

sorry, context window not content

63 in the end

Took 43 to get a green

Solved the new semantle, #19, in 2 minutes, 36 guesses

got it in 61. trimming the fat a little, but once again, I could have saved ten guesses if I pivoted earlier once I knew I was heading in the right direction.

I solved Semantle #19 in 41 guesses. Most straightforward experience yet, still fun. First green was something crazy high like #969.

I have now three other friends playing this

26. far and away my best result, and looking at others' results, this seems a particularly easy one.

lol every time I'm like "damn, awesome, I'm getting better at this", I look and everybody else got it in like 10 guesses

Oh don't worry I gave up on today's.

51, my best showing yet. I think the game is getting a little easier, altho practicing no doubt helps.

50-100 guesses is still a big commitment for most people (a coworker said I must be spending all day playing if I was guessing hundreds of words) so I feel like the current level strikes a good balance between being hard enough to be challenging but not so hard that it’s a chore for die-hards only.

didn't get yesterday's but don't feel that bad about it - for a while my only green was electricity. I tried water, gas, utilities and they were all cold. shrug.

I've started to use a category approach, like....if I start to see my results improve, even if cold, I have really started to strain my brain to figure out if there's any linkage between them.

for instance, tonight's was consume, and pretty early on, I was in the 20% range, so still cold, but on the right track, and it was all food related guesses like chips, mashed, broccoli, burger. so originally I was guessing all food stuff, and then I said 'well obviously if it was a type of food, these wouldn't all be 20%s', so I started guessing words like meal, snack, which were also....in that range, but slightly higher.. And then I realized well shit, it's probably not about food, or the meal itself, but how it is ingested and within 5 guesses I had it. so the 'waste' was all the pointless 'food' guesses.

fuck shit ass shobadeedoowop

food related guesses would have definitely helped in my case, all my warm guesses were destruction-based e.g. burn, destroy, drown so struggled with words like murder, kill, crash and accident until i got hungry and punched in 'ate' and from there i got it immediately.

170 today, but i think i'm getting the hang of it somewhat

I solved Semantle #19 in 33 guesses. https://semantle.novalis.org

I'm improving.

30 but could have been quicker! First green was guess 15 with drink/981 then i messed around too long with synonyms for that (verb and noun) instead of trying related concepts.

Today's, #19 - 21 guesses.
Was about to sack off this game but now I'm back in!

49. Had 997 on my 20th guess but got fed up of firing related guesses with no luck so resorted to a thesaurus.

Today's word is v close to a word that I got stuck on yesterday (but which barely had any relation to yesterday's word) so it was in my mind

This game reminds of that scene in Seinfeld where Kramer tries to guess Georges bank card password "Bosco". He straight away goes in for the food area and quickly hones in on chocolate. Then chases George out the apartment yelling all the wrong chocolate products.

Gave up on 200, got within 178 early on but every other guess was cold.

Got it in 60 with only one guess in the 900s. I've been too focused the last few days on related words (synonyms) instead of the underlying process, so that change helped.

yeah I got super caught up on a specific vector of do - manage - require etc

Got this one in 19, but very lucky again

In 16 rather. I had a bunch of cold guesses (26 was the highest) and then I randomly hit on a super close word that made it easy.

Sigh. I just spoiled today's puzzle (#19) for myself because I un-hid Neando's post thinking it was about yesterday's. Using the puzzle number is more useful than I had realized because it changes in the evening during prime "working on stuff" time.

The page also shows yesterdays word, so be weary of that too

But you can't play previous days' games.

I think someone upthread said they'd found a way of doing it

or i'm telling big fat fibs

Oh maybe you can do it by changing the system clock idk

oh that could have been it, I guess they wouldn't see the new page with the answer then

Yeah, if you leave the tab open, it's still operational the next day.

212 today. Got to 997/1000 very early, pretty much at random, but was struggling for a bit to nail it.

#19 took me 193. I genuinely thought I was going to get a really low score, but it turned out my green guesses were related to creation (PRODUCE, DISTRIBUTE, SYNTHESIZE, GENERATE), when I needed things more related to destruction (DIGEST, EAT). The key to my turn was actually scoring high with DEFECATE, hehehe. Got it fast after that.

8th guess botches!

16 on puzzle #19!

My best yet and feels like less of a fluke than my previous best, where after getting 15-20% on "person" I typed in a random profession which turned out to be half-connected to the answer. (Even then I spent a while going through the wrong half of the concept and guessing other related professions.)

51 today

a relative saunter

This one (#20) took me 117, but the solution actually popped into my mind much earlier and then popped out again, so I convinced myself that I had already guessed it and didn't try it for ages.

#20 took me 74 - got to green relatively quickly and then spent an absolute fuckload of time (and probably 40+ attempts) racking my brain for Perry Mason words - there are an awful lot of 'em!

fuck fuck fuck.

got it in 143. but I was...once again, in the right ballpark by like 80. I went through every courtroom related word imaginable too.

I solved Semantle #20 in 117 guesses. https://semantle.novalis.org/
Guess 23 was sentence which got me to the top 10, then I futzed around trying every related word I could think of.

This was a weird one for me because I only had three guesses in the green before I solved it, but I had more than 100 completely cold guesses.

it was the opposite for me!

I solved it in 70 but 38 were green guesses - it was pretty clear what the ballpark was, just a matter of hitting the right term

ok so i have a query and maybe someone with word2vec smarts can help me out and sorry if it's totally dumb - but in #20 the trail leading to the target word was very courtroom-based, which is fine - but in this case the word itself isn't exclusively a "court" word - like, it's a word with general meaning that i guess is used most commonly in a courtroom context, esp in media - but words like "implore" and "beg" i would consider very similar in meaning, but these are way down the list - does that mean that being 'usage adjacent' is more significant to the dataset than being 'meaning adjacent'?