Is WORDLE with longer words easier or harder?

[@Psychomaniac14]

I like how satire is the best 6-letter starting word

[@adamel-sawaf4045]

I really enjoy how you explain your code and your findings! I think the one thing you should do more though is analyze *why* you think the results turned out that way. For example, maybe Wordle is computationally easier with longer words because the player gains more information per guess, proportional to the total amount of possible information (total number of words with that many letters). That could explain why 1 or 2 letters are actually easier than 3 letters, because there are so few words in English that are that short, that the “word length to total number of words” ratio increased. I wonder if this ratio would follow the same trend as the difficulty that you found. This value can maybe approximate the average number of guesses needed to guarantee a win? But anyway, keep up the great content man. ❤

[@DanielBerke]

For one-letter Wordl, "O" is sometimes used as a single word in older written dialog, such as Shakespeare's plays for instance. Nowadays it's usually spelled "Oh", but I think it might count as a valid (if archaic) word. Still wouldn't be a very interesting game, though. 😄 Also, wow, as a native English speaker with what I like to think is a larger-than-average vocabulary from a lifetime of curious reading…I didn't recognize _so many_ of those words. Still got a lot left to learn, I guess!

[@Liggliluff]

I think place names was used, in that case E and Y are valid place names, there are more.

[@448285]

"to simply come up with a word that is...eleven letters is a serious challenge in itself. It's IMPLAUSIBLE to expect your brain to MANUFACTURE them ARBITRARILY and CONFIDENTLY." Well played.

[@allideni836]

I love your videos! As useful as deep-learning is, it's super interesting to see algorithmic solutions like these.

[@GamesComputersPlay]

Thanks! To be honest I don’t see machine learning and algorithms as necessary mutually exclusive approaches. There are tasks done better by algo, and those done better by machine learning. There is a lot to do with the available data, an opportunity to run millions of iterations and so on. The best result, I would suggest, is when you break the task into separate parts and use both algo and machine learning together - giving each an opportunity to do what it does best. I am looking forward to dipping my toes into machine learning - and I am excited to learn and use it in my projects. Although it is rather a midnight hobby for me, it takes quite a lot of time.

[@joshuasims5421]

I think the result makes sense, but you could model the difficulty of high-length games using word frequency. Even as a rough measure, you could find the product of the relative frequency of each word used in a solution, and then assign a difficulty score, assuming that infrequent words are harder to remember. Then you could find the average difficulty score and probably see an increase in difficulty in higher length words.

[@jammesdean]

Fantastic video as always! It’s really cool to see you applying an old bot in a new way! Keep up the fantastic work!

[@landonkryger]

I'm curious how big the word lists are for each letter size. Obviously, the long words have more opportunities to give you clues, but surely there's a lot fewer of them to guess too right?

[@GamesComputersPlay]

You are quite right. The number goes down dramatically, especially of secret words. Too many of them are just obscure professional terms. (I am guessing this is the reason 12-15 words are excluded from Hello Wordle). Here's teh stats (2 and 3: secrets filtered by me, the rest is from Hello Wordle's source code). 2: 53 secrets, 101 guesses 3: 458 secrets, 1015 guesses 4: 2351 secrets, 4030 guesses 5: 3242 secrets, 12970 guesses 6: 4562 secrets, 15786 guesses 7: 5137 secrets, 24028 guesses 8: 5328 secrets, 29765 guesses 9: 5056 secrets, 29150 guesses 10: 4030 secrets, 22326 guesses 11: 3022 secrets, 16165 guesses 12: 2055 secrets, 11417 guesses 13: 1438 secrets, 7750 guesses 14: 751 secrets, 5059 guesses 15: 434 secrets, 3157 guesses

[@heitortremor]

I love hello wordl! I was so happy when I found a nice open source Wordle clone!

[@joshuasims5421]

Another variation to test: how does the computer fare if it doesn’t know how long the word is? So you can guess words of any length and the answer may be of any length. How would a human fare in such a game?

[@GamesComputersPlay]

Interesting idea. I haven't thought this one through much, but I think a lot depends on how exactly you implement missing letters. Is it done with a 27th separate symbol, say “!”? is the word always starting from position one and padded with those “!” at the end - or is it randomly anywhere in the middle. Is the player allowed to use “!” as a guess? If so, I think all “!” is the best first guess to rule out all wrong lengths.

[@CamerTheDragon]

(Spoilers for vid) Cool vid, I've never played wordle (altho I've seen it in vids before like yours and 3blue1brown's) but I remember thinking about this before and I guessed the same as what the computer got, since with fewer letters you get less info about the secret word and with 6+ letters you get more information than usual, plus once you get to double digit letter words there are not very many words up there comparatively to 5 letters and such so there are not many possibilities at all after a guess or two

[@Oliisawesome]

I think I’d be really good at wordle if the words all had 26 letters

[@AbelShields]

I think the reason harder difficulty makes the bot perform better is because it restricts weird words - I saw one (was it 11 letter?) guess starting with a double o, that seemed like a bad pick - I think this was on the double h one it got wrong. I believe if this was done with a perfect bot based on entropy, the harder levels would indeed make it tougher to win.

[@GamesComputersPlay]

I think you are right. Since the bot is a rather weak one - choosing 5 random words will usually leave you with some low-quality guesses, you just pick “the best from the worst”. High difficulty in this case is a filter that filters out more of that randomness, leaving more or less relevant guesses. My guess it would not be the case with a more powerful bot (the one that uses, for example 100 or 500 random words to pick a guess).

[@jakubkubiak2671]

After your two previous videos, I still think about one thing. Using the "Color + weight" method, how do you choose between words consisting of the same letters? To be more specific, how do you choose between words having the same score (according to their weight)?

[@GamesComputersPlay]

I’m trying to think back, I think I used the weight of a “unique letter”. So if you have 2 As in the word, only one gets counted. Two reason s for that: - Using words with repeating letters is considered a “bad move” even for human players. You’d get information only about 4 letters, while you could have gotten about 5. So the algorithm should be discouraged from using those, unless they really outweigh other words, even with these limitations. - When I actually tested an approach where you counted weights of all letters, including repeating ones, the best opening guess was “SASSY”, (or something to that effect , with 3 of the same weighty letters). Which is clearly a terrible guess.

[@jakubkubiak2671]

​@@GamesComputersPlay I understand the logic behind the repeating letters. My question was about something different. What I was asking about was the situation when the list of words, from which you have to choose the guess, includes e.g. two words - "soare" and "arose". These words consist of the same letters, so they should have the same score (if I have properly grasped the idea of counting scores). In that case, how do you choose between them? Which one is "better" than another and should be our next guess?

[@GamesComputersPlay]

@@jakubkubiak2671 Ah, I see now. In the version used in the first video these two would have exactly the same weight - I think the word would be chosen randomly between highest scores, if I remember correctly. However, there was a second version, (I briefly talk about in the second video) where I calculted weights for each position separately. In that case A as the first position would have a different weight from the A in the third. If the total sum of 5 weights would end up being the same, however, it would be down to random generator again.

[@jakubkubiak2671]

@@GamesComputersPlay Ohh yeah, that is what I was talking about. Thank you for your responses :)

[@Liggliluff]

What I've been curious about now is wordle in Swedish, since Swedish has the extra letters Å Ä Ö and É. With a bigger alphabet, it should be harder? I do also wonder what the best first word in a 5-letter game is.

[@GamesComputersPlay]

My intuition tells me it is going to be a little harder, but there are other factors too - how many words are there, how frequently various letters are used, how letters are distributed within words / within a certain position of the words, how many repeating patterns are there. Mathematically, I think it's a bit too complex of a question to come up with a simple answer - but a simulation can. In real life it also depends on an actual person. I don't think you can know 2 languages with exactly the same level of fluency - one will always be a little ahead, giving the player a better chance.

[@aydenherold443]

Yeah but... what about 0 letter worldle...

[@YoutuberFellow]

where the best first guess is ""

[@Liggliluff]

Arbitrary limitations are no fun. That website has 15-letter words, so why isn't it normally available?

[@GamesComputersPlay]

It distorts an interface a little - maybe that. Or the vocabulary is a little limited. Or it makes the game too hard. No idea, to be honest.

[@keejay98195]

4:44 why would the bot choose a word with 3 O’s as the second last guess when it’s already ruled out 🤣 also I wonder how much better this bot could perform if you would allow non-words just to get information about frequent letters

[@GamesComputersPlay]

That's the thing - the algorithm does not limit itself to only the words that fit the current restrictions. It tries to find a word that narrows down the remaining possibilities the most - and sometimes it is a word that is completely out of line with the guesses.

[@lel7531]

You should do a solver for a connect4 but with 5 in a row instead

[@theregularterrarian3933]

I bet there are like 10 15 letter words tops

[@GamesComputersPlay]

434 secrets, 3157 guesses. However, you do have a point as it is full of such gems as polynucleotides, connoisseurship, disarticulation and ecclesiological. And this is from the curated list of secret words, you can imagine what's going on in the "allows guesses" list.

[@iskanderaga-ali3353]

Hello mate, can you remind me of what happened on the Tiananmen square on may 20 1989?

[@GamesComputersPlay]

Technically, not much. Maybe you wanted to ask about 4th of June of that year?

[@arn0000]

first time i'm writing 'first' but i do hope the vid's like the other ones, interesting/entertaining :P

[@swordfishxd-]

6th

[@qo7052]

seventh