The Pretzel Conundrum
Вставка
- Опубліковано 18 тра 2024
- This is a video about pretzels. Or seltzer. Actually, it's about Scrabble.
More precisely, it's about a small handful of similar words that create quite the conundrum when starting a Scrabble game. Even the strongest Scrabble players in the world disagree over how to approach the Pretzel Conundrum!
Huge thank you to:
Josh Sokol @axcertypo for interview footage
Morris Greenberg for Python / Plotly wizardry
Cesar Del Solar of Woogles.io for BestBot play recommendations
All of the listed champion / grandmaster players for their generous insights
Sometime soon, I plan to resume streaming on my Twitch channel:
/ wanderer15
I will also soon resume my weekly show on the official Scrabble Twitch channel (Tuesdays 3-5 PM ET): / scrabble
Play Scrabble at playscrabble.com! - Ігри
Beside the Safe option and the Bold option there's a third option, where you never see the bingo to begin with. That's the one I'm most likely to take.
Yep, it's a "tree" for me.
I'd have been very proud to find Repel
Except now that I've seen the word discussed at length, I'm much more likely to spot it.
Maximum psychological advantage is playing the bold placement vertically
the dude at the end did exactly that lmao, no wonder hes the champ
That's the most funny thing in Scrabble -- it doesn't matter at all technically, but humans are humans...
I think I would play TREE. :-)
I love josh's answer. Games are meant to be fun, so fuck it, we ball
Came here to say something similar, stayed to agree
The 3-axis scatter plot was perfect and so clearly explained. ✅✅✅
Why am I not surprised Josh would go for the bold option on both?
Josh has that dwag in him
your scrabble coverage is by far the best coverage of any competetive game or sport i have ever seen.
100%! Most of us don't even play Scrabble and we're still hooked
I am a chess master and this man makes me appreciate Scrabble way more than chess
@@degenerate82 chess has extremely good analysis too
It's rare to have a channel that both analyzes games expertly while also playing that same game expertly too. Not to mention the regular uploads and quality editing and guest appearances. Really high quality stuff.
Thank you, that’s really too kind!
Can we get this channel large enough that when Will goes to ask his friends their thoughts he can call up Nigel Richards?
I would love to see more of these Theory based videos. Theyre so interesting to see the psychology behind masterful Scrabble plays!
I feel really confident about how I approach this conundrum. The better my opponent is, the more likely I take the bold approach. For the same reason, I’m more likely to open triple lanes against higher rated players, who are likely to score well irrespective of the lane.
The other consideration is where I stand in the tournament. In one Nationals, I had a decent record but a poor spread on Day 4, so I opted for the bold opening because it seemed like a necessary risk to come back. It didn’t work- opponent got there first-but I don’t regret it.
I agree with you 100% and was about to say the same thing. If I were playing, say, Joel Sherman, I know that starting with a 92-pt lead (or even a 106-pt lead) isn't "safe" but is definitely a big plus. And then if he does NOT have an S or blank, he's likely to have to play defensively. Add to that the likelihood of picking up an S or blank myself, and it means that I'm very likely to win that game. Of course, if he has one of those hooks, he'll likely score big and go on to win... but that was going to happen anyway!
It would be interesting to see a Scrabble engine that takes this into account. It would suggest more aggressive tactics if your opponent has a higher ranking
@@QuarkTwain In the world of chess engines, this is called "contempt," and it usually comes up in the context of making or avoiding a draw (high-level chess is a very drawish game). When the engine has high contempt, it is less willing to accept a draw in a position that's otherwise defensible. Effectively, this makes the engine play more aggressively, at least in the middle game.
It's like how NFL teams are more likely to go for it on 4th down late in the season/late in the game trailing
I prefer the bold placements generally and feel very validated by Josh's comments 😀
I'd say what the current engines say is almost useless here. As advanced as Macondo is, it doesn't yet have the tools at all to correctly evaluate such a position.
For a few of the front hooks, there's a small added possibility that the opponent might not even notice them. Generally very unlikely but adds a few % to the value. Similarly, there might be some words where the back hook is uncertain enough that the bold placement might leave the opponent unsure if the word takes a back hook... especially under double challenge.
The evaluation is just a simulation of win%. All the engines are saying is that from simulating games out, the number of times your initial lead is overcome by the end of the game is more in one case or the other. This isn't useless because while not literally definitive, this evaluation is significantly better than any evaluation a human can do on the same subject, and why the engines are able to tease out the minute difference between a word that is good to play aggressively vs not.
@@Ecrilon I guess useless is a strong word, but I'd expect the simulation results to be highly misleading in a situation like this because the moves played within the sim don't match the moves that would typically actually be played in a simulation like this. So it's a case of the "garbage in, garbage out" principle
@AlexDings I am pretty sure simulations are overly generous on the bold plays, actually, not as much because human play won't mimic computer play, but because at the end of each iteration of a simulation, the computer will give an estimated win percentage of the position based on score differential and tiles remaining, without factoring in board dynamics. In fact, the fewer plies you use, the more the computer likes the bold play, since the hook gets cashed in less often and the win percentage estimate will be too generous given the dangerous hook out (try PRETZEL on 2 vs 4 vs 6 ply)
@@Ecrilon But the problem is that the engines simulate the games using what they consider to be optimal in the subsequent moves, and that can have its own bias and issues, exactly in tricky, double-edged situations like this.
Pretzel Logic
Great album perfect to listen to while playing my favourite game! Great video
Bold move: opening with H8 PRETZEL
Funny move: opening with B8 HITLESS
Turn 2: playing A8 SCARED
Alternative turn 2: A2 FUCKERS
This video was great. Well researched, original research to boot, great editing, interesting thesis, and much more high effort in general than just analyzing a competitive game.
I always click 'like' as soon as I open Will's scrabble videos as I know they're going to be so good. I would take the bold option against a much weaker opponent as I want to decrease the chance they can beat me on lucky draws. I would take the bold option against a much stronger player for the same reason, increasing my chance of winning from a lucky draw. Against any player near my level I would play the safe option.
I don't know if you've heard, but there's a new Scrabble TV game show coming soon, hosted by Raven-Symone, and it will be much closer to the board game than the '80s version.
This is a good opportunity to do a video on the '80s version that you and I loved as kids.
Didn’t know who the host was, cool!
7:27 I love this. Playing a move that might be technically suboptimal to get into your opponent's head, and because it's fun. Who says competitive play takes the fun out of games?
Especially considering tournament play is time boxed. Even in casual games here on UA-cam, I've seen pros get scuppered by agonizing too long over close calls, leaving them with no time in the end game.
Its really cool to hear scrabble players talk about words as more or less "defendable" or "risky". This honestly feels a bit like getting into a fighting game, where yeah I might understand the basic game, but its awe inspiring to watch pros pull off cool plays.
I’ve made a couple videos now comparing Scrabble to other games and I really want to do one someday comparing it to fighting games (I have a couple specific ones in mind)!
Minute or so in I was thinking the aggressive approach looked like a Josh Sokol type of play! Such an incisive position, and explanation and editing of your video is excellent.
I haven't played scrabble in forever. I think I got recommended this channel because I play other games competitively and I watch chess videos and stuff. I would play the bold approach against players that I feel are stronger than me, and the safe option against opponents I think are weaker than me. Which menas, I would always play the bold option, but I'm imagining I could get good :p
Thanks for stopping by and giving my video a try! Your instincts definitely mirror my own.
I once played DEUTZIA aggressively in a tournament. My opponent didn't have the correct tiles for a hook, so was forced to make his best move, and leave the hook up. I picked up an S, so followed up with PLAYS/DEUTZIAS, and was able to take a huge lead on turn 2.
Very cool premise for this video! I love the scrabble histories as well but this dive into a very specific conundrum is fascinating. Thanks Will
this is my favorite video of yours yet; I was even thinking about this myself with PRETZEL just the other night and I (a scrub) decided that the safe option had to be right. I had no idea that there was such an analysis of words like Pretzel, so this video really opened my eyes to the level of analysis that has gone into a very specific, and very cool, scenario.
I really like these conceptual based videos! Amazing entry point for casual players to see optimizations to mindset not memorizations!
LMAO Im actually proud that I realised what the video was going to be about after like 5 seconds, I officially now scrabble stuff yay
While I'm not a Scrabble grandmaster yet, the upside to playing offensively for a score difference of only 14 points isn't worth the stress (additional fun element aside). Tournaments are stressful enough, and it's important to preserve energy whenever you can. I'd rather take a small point sacrifice and move on to the next rack.
I feel like in many ways the 'bold' placements are less stressful, no? You're going to have a lot of positions where the spot is either hammered immediately or blocked immediately - which lead to a normal game - and if the spot stays open for a prolonged period of time, then it's all in the hands of the tile bag. I feel like most of the stress is placed on the opponent rather than on you. I understand your point of view though
@@AmaranthRBY I'm sure it's fun sometimes to take a punt and hope for the best. The only scenario I can think of where it would make sense to play bold is when you're required to win by a huge spread, so creating more volatility is the way to go. Otherwise, I'm happy with the regular pretzel :)
absolutely loved this video and would love to see more things like it about scrabble theory as opposed to a specific game or situation overview
This video is going to get me in trouble at my next tournament, isn't it
Loved this video!! Not sure if The Algorithm would agree, but I would be super excited to see more uploads like this one, breaking down strategy and not focused on any specific real-life game/player.
Thanks! I do pay attention to The Algo but I also run with the creative spark any time I feel inspired by something (which happened here basically at random).
And among all the brilliant points about correct play, he also just sneaks in a bombshell insight at 6:23 that I'd never thought about; that looking at how the strongest players play, is not a plain measurement of what is correct, it's a skewed measurement because strong players more often play opponents weaker than themselves, and adapt to that situation.
It’s true. Several respondents said they would consider playing the aggressive move only against other elite players (though many said they would never play boldly against any opponent ever).
Omg that's an incredible video. Thanks Will!
Very nice video. Thought-provoking and informative.
Love this new style of video!
Fantastic video - thorough analysis.
I can think of a quick analogy in poker called "implied odds" and "reverse implied odds."
Low pocket pairs like 22 can be valuable in poker as a way to "set mine," fishing for the roughly 1 in 9 chance of another 2 coming on the flop giving you at least 3 of a kind.
This works because low pocket pairs have good implied odds. When they lose, they tend to lose small and fold early, as you're not likely to invest much with only a low pair. When they hit, they tend to win big, as you'll have a very strong disguised hand that other strong hands (top pair, overpairs, two pairs) might not expect. Even against straights and flushes you have outs to a full house.
However, there is something to be wary about, especially with the lower pocket pairs (22-66)-- reverse implied odds. It's the same concept but one level deeper, and applies when both players have more money in front of them.
The problem with low pocket pairs is that although they do well against strong hands, they can get dominated when a higher pocket pair also hits a set ("set over set"). When this happens, you are almost drawing dead most of the time, needing to spike 4 of a kind to win. Hence, when players have very deep stacks in front of them, low pocket pairs become less valuable because you might not be able to fold them yourself to an enormous bet, even when dominated.
I always love seeing a retinas reference. Great content. Thanks!
wanderer15 true fan alert!
One of my favorite Scrabble videos.
Count me in team Sokol, and imo it's always correct to go bold against higher rated opponents. So, if I'm playing Nigel, Will, or any of the experts polled here I put them to the sword, even with BEZIQUE!
Dude this reminds me so much of when I was learning poker. You just have to get used to bad beats and accept that you're playing a sample size of many thousands of games instead of just one. Once you get in that headspace 62% chance to win or at least get a dominant lead is a no brainer. That being said, I imagine it'd be easier to live with the results if you were playing like a scrabble ladder or something vs a one off tournament. I guess that's where the scrabble version of ICM would come into play.
maybe scrabble players that also play poker have this mindset in general
It's not quite the same in my mind for 2 reasons: (1) both the bold and non-bold play put you above 62% chance to win (probably even higher, more like 70%), it's just if the bold works well for you it jumps to 99%, whereas working badly puts it at 40%, while the safer placement has you hovering in the 70% zone for longer, and (2) poker is a game where many hands are played and the objective isn't to win the most hands but to win the most money, so you should base your strategy on the price for playing the pot compared to relative odds, even in the cases of bad beats. Scrabble tournaments are based on wins first and foremost, spread as a tiebreaker. I think you see a lot of the top players favor the safe positions because they trust they can convert a 70% computer win probability in human play more often than the weighted average between the 40% and 99%.
I feel like the bold option is more fun and also (obviously) higher variance which plays to the underdog's favor. I am so lucky to have gotten to play against Josh and our game was an absolutely *ridiculous* high scoring slugfest, with me benefitting a lot as the underdog from his fearlessness at leaving scoring spots open. Part of me hopes you someday cover that game on this channel!!
That actually would be a great one to cover. It’s still lurking on UA-cam, right? (Scrabble Go channel as a VOD maybe?)
I had the pretzel conundrum in the form of RELAXIN during a club game some time ago. I definitely took the bold move for the reasons Josh Sokol stated -- more fun. Easy enough to do in a club game. In a tournament where it mattered, I would likely have skewed toward the safe play if I was in contention. But with this analysis, I feel equipped to make a much more informed decision should it happen to me again.
I will almost always go SAFE, I don't have the bravery for the bold moves unless I get desperate 🤣🤣🤣
What will I do next time I am lucky enough to draw a pretzel?
I won't even see it.
So LAZIEST is a safer option right?
HastyBot: immediately gets a g in a starting rack
What a great breakdown of scrabble game theory. I liked how you showcased the nuance of the plays and asked others for their opinion because you're not sure yourself. The way I see it, if you follow the mathematically correct line and then play a million games, you'll win the most often. But if you only need to win one tournament game, you can definitely play it safe and focus on early leads. Will, you're kinda the Kendrick Lamar of football
Better than the Drake of football. Thanks!
Interesting analysis! As a strong intermediate / low expert player myself, I will usually play the risky version if I think I am an underdog against a very strong opponent. I might need some luck in any case, so I might as well hang the game in the balance right away.
Such an amazing video!!!!
Another element (not needed at the highest level 😊) is whether your opponent is likely to know the hook. I played BRONZER years ago in the ‘risky’ position, then was later first to pluralise. So satisfying 😅🎉
Very nice!
i think in tournament play, it depends on who you are facing as well. if i know im better than my opponent, i will be more inclined to try a riskier play. its a good way to further increase my spread and even if he takes advantage of the TW score, i will still be able to beat him most of the time.
My favourite word in scrabble
Ha, glad you enjoyed this pretzel content!
Great video!
I wonder how much of the calculation on SELTZER should consider that by playing the risky option, in addition to the -S hook, it also puts the S in SELTZER below a triple word, making any bingos the opponent may have that can take an -S that much more powerful. The score gap for SELTZER is a few more points than PRETZEL, but to me it looks like there's additional risk involved, just not risk in the opponent playing an S. But maybe I'm just overvaluing that possibility.
That said I would still 100% take the risky play. In my experience, playing aggressive is just more fun.
That’s actually a great point. However, keep in mind that even safe SELTZER gives up all the same bingos as aggressive SELTZER - if all the tiles are low pointers, the difference may only be a handful of points. If you do happen to hit a high-pointer in the right spot, the triple-word S bingos will hit harder for sure - but even then the gap between those plays will always be significantly less than the gap between the big hooks and normal hooks.
The way my mother and I play (when we have the chance to) is quite similar to this, as we always go for the highest scoring plays to maximize our *combined* points. It adds a layer to the game where you're actively pushing where you normally wouldn't, it's great!
Cool variant!
I find the math and psychological pressure being quite convincing as reasons to play the aggressive option, especially as someone who is not usually a favorite to win Scrabble games.
Wow scrabble strategy sure has a lot of twists! Kind of like a garlic knot or braided babka!
I'd say if you're playing someone higher rated than you you should play PRETZEL, SELTZER, RELAXIN, etc in the "content" spot, because you're already an underdog to begin with, so if it's close to 50/50 whether you're going to get the hook you need first, which I think it is close to 50/50 in most of these situations, it's often better just to play it against a higher rated opponent or an opponent you have a bad record against. If you do hit the spot before your opponent and your opponent is higher rated, the handicap of being 150-200 points behind is often enough that you can just play at your normal strength and still win comfortably.
Josh Sokol mentioned at a certain point not playing for fun.
Reminds me of a Smash Bros melee player named Amsa, who was known to say "i play to win"
As a mediocre Scrabble player with poor dictionary word knowledge, I think players like me should ALWAYS go aggressive when we get the chance, especially against stronger opponents. If it comes down to word knowledge, I'd lose to a pro player every time. But leading with aggressive plays like that can increase the chance of winning to almost a coin flip.
Strong players facing a weaker opponent probably shouldn't do that, because high point hooks are very easy even for novice players to find in these cases. But, maybe if you're a competitive player facing another player of equal or higher ranking, it pays to be as aggressive as possible.
This makes me want a scrabble anime lol. Champions monologuing their moves haha
Letting someone triple off your z evens the game. Tripling off your own z? Absolutely crushing
I'm 100% with Josh here. I don't play scrabble, but I'd always go for the most fun, and often rewarding play. Even if they manage to use the open spot for a retaliation, we are even and have a game on our hands. If they don't, I'm very much in the driver's seat.
If you allow smaller tiles being the one under consideration, I've heard a horrifying example: a player putting the M on the double letter score for OOMIACK and his opponent had a W for WOOMIACK which is also an acceptable spelling.
First, I agree with Josh that it's more fun leaving the spot open. It's just more interesting doing it that way.
Second, if I'm playing against a weaker opponent, they probably won't see it. (I'm not that good at the game, and everyone I play with is twenty years out of practice.) If I'm playing a stronger opponent, then I need any advantage I can get, and a 200-point lead is definitely a good advantage.
Hi Will. Is professional scrabble always played 1 v 1? Have there ever been any classic games with 3 or 4 players all playing against each other?
It is indeed always 1v1 though I’d love to see other variants popularized!
I guess the correct answer is "it depends". It depends of parameters that are not easy to establish, and beyond the mathematical study proposed. It depends of your opponent : is he enough good to give him 130 points? does he seems unconfortable with his first draw? factors that can influence your choice are also how you tournament is going? did you have misfortunes yet? or does everything goes well and you feel strong and invincible? Matsumoto's point around Entropy is key. Know yourself and play accordingly to what makes you more confortable than your opponent
Excellent analysis. I’m a ‘bold-y’ btw :)
5:12 I'm curious about the engine's bold preference for LALIQUE but coin flip for SILIQUE, two very similar words which makes it easier to compare them. I would have thought the engine favours bold when risk is lower, and SILIQUE has only 5 back hooks compared to LALIQUE's 6. Do you have any idea why?
My best guess is the risk of putting S of SILIQUE on centre square factors into the equation, but I wouldn't have thought that it would tilt the risk that much. I'm not sure if you have more insights on this.
This is probably just a bit of noise caused my lack of thoroughness on my part. When I run the sims for these two again, they’re both in the coin flip category for me.
I'll take my sanity above the 14-18 points, thank you very much! I cannot imagine playing the "bold" option in a real tournament game... I can only imagine your coach or the commentators/observers laying into you after the game for "throwing" after getting such a heavenly opening rack and going on to lose.
Of course, if you actually went second and the first turn your opponent exchanged 3 tiles for example, playing the "risky" option becomes just plain suicidal.
As a “safe” option picker myself, I totally get this. I don’t know if I’ll actually have the stones to try a bold placement if I draw a rack like this…we’ll see!
Well that is an interesting premise.
I'm actually quite curious about the zinnia vs nib question at 2:00. What would you favor there and what do computers say?
Computers emphatically consider NIB a mistake - the player trailing likely needs to leave the slot open and hope for the best there. But again, against a much weaker player, I might be inclined to block and battle my way back in a game where my opponent has much more difficult plays to spot than giant S-hooks.
@@wanderer15 That makes sense. NIB also does not completely block the hook. It would be devastating to play NIB and then see your opponent bingo with ASBESTOS.
Hi Will, is there a link to the 3D plot used in this video? I would like to dive deep into the chart
Yes, I’ll try to remember to dig it up and post it here
every time I get aggressive I'm seemingly always punished instantly for it no matter how often the probabilities seem to be in my favor. So I'll just play it safe.
Any chance you could pin a comment asking the two people that chose safe placement for pretzel and bold for seltzer why they were inclined to play differently for each? I'm curious. My inclination is that it's just point difference but I still think an answer from a top player would still be interesting.
Maybe because there is a lower probability that your opponent has drawn an S with seltzer (only 3 S's left, so be bold) than pretzel (4 S's left, so play it safer).
The reasons Adam and Jackson articulated to me align with my explanation of “what the engines want to see” from a play like this. Their feelings were that for the right price (high score differential and less immediate risk), there would be a point at which PRETZEL-like plays become acceptable, and that point lies somewhere between PRETZEL and SELTZER.
When I was playing at a higher level, especially when I was mainly playing CSW against North American opponents who had not yet learned as many of the CSW-only words as I had, I was very risk-averse with these racks, preferring to coast to likely victory without setting up a huge coin flip. Now that the standard of play has increased while I've gotten much rustier, I'm more inclined to take the points and I'd probably play 8h SELTZER against anyone 1800+
I find it really interesting that only one of the GMs you asked had a different answer for seltzer than for pretzel, when both engines had a different answer
Actually unless I edited it wrongly it was both Adam Logan and Jackson Smylie saying that. Still very much a minority answer, but I think it’s actually the closest to my own answer now.
I LOVE PRETZELS 🥨🥨🥨
I love your videos !!!
YEAHHH
I love both Pretzels and your videos
I love pretzels and all of your nice comments!
At my level I always go bold but not because of any psychological advantage, more that it will be a long time before I hit a score over or close to 100 again. I'd like to see a video on highest scrabble openings in professional scrabble if you haven't already done one.
I will play safe Will. I just learnt PRETZEL. Thank you Will.
Tournament, I'd play Safe... games with nothing at stake (i.e. club, friends), I'd play Bold...
Just by "Peel" to avoid this problem.
Hey Will, out of curiosity, does your Scrabble expertise translate well into Wheel of Fortune or Hangman? Like do you just destroy those games when watching/playing just because they're also about words and letters?
I would not say I’m especially good at those games - I’m solid, but not godlike. Boggle, on the other hand, I have gotten very, very good at - arguably better than I am at Scrabble.
Scrabble game leveled up!
I wouldn't even see I had Pretzel to begin with
They don't say "fortune favors the safe."
i love pretzel graph.
Scrabble theory! Pretty cool.
I would go for the bold play, but that's simply because I am not good at scrabble and I just go for the high scores. In games with players of a lower level like myself, I think it's better to go for bold plays because the opponent lacks the skills to block or respond well, or just in general finding good scrabble words. Thus, I think you'll benefit even more, even though engine analysis might say otherwise. And as we all know, lower level players are not engines so very often won't find the engine-like plays.
But this is all just a thought of terrible scrabble player, aka me.
Have you searched for examples of tournament games where players opened with a PRETZEL?
Yes, there are a handful! Too expert Matthew O’Connor told me he opened with GENIZOT# in the aggressive spot - when his opponent didn’t even consider challenging the play, he then hooked it with an S (it only hooks with an H) and got away with a 90 point phony play.
Always risk it for the biscuit
Great analysis. Now do 'EHIORT?' going second, opponent opened with 8G ZED. I'm curious what the current strategy is.
I always chose the bold option when I was a child scrabble prodigy.
Are there any Nigel games on record of him having to choose safe vs bold like this?
I wondered the same thing after watching this and went through his games - doesn't seem like there is one on record. Otherwise we'd know the correct answer 🤣
not quite... the closest would be him going 4 PRENZIE 4 max pts on stream ― no hooks thrown out (surprisingly) 👀
Great video. I think if you are up against a much weaker opponent you take the low variance route in most cases, and play for maximum EV (expected value) against a stronger player
I tend to agree with this. Thanks!
One question: Pretzel only comes up q out of 4750 times (I'm assuming this counts the blank).
What about if we account for all 100-or-so words that fit the conundrum criteria? If it's common enough, sounds pretty worthy learning for a player.
It actually does not count the blank, because I couldn’t figure out a very easy way to compute those possibilities quickly. With the blank factored in, these situations become significantly more likely (though still quite rare). (edit: sorry, I misunderstood - *all* of those words together come up once per 4750 games. because they all have a rare tile in them, it's very difficult to draw them)
Seems those words constituted by a 10 pointer and all 1 pointer else would work quite a lot
What did you use to make the 3D graph?
Another expert player, Morris Greenberg, helped out with his R knowledge and Plotly!
math, psychology, bingos. how can you not be romantic about scrabble
How likely are you to draw a PRETZEL if you start with the best exchange move and your opponent exchanges as well?
How is the engine considering those factors when recommending course of action?
It's essentially never blocking the way a risk-averse human might - Scrabble simulations automatically assume players will make a play with the best combination of points scored and good letters left over. Usually, this is true, but this situation is highly unique with such a high motivation for both players to block the new spot. However, it's hard to say which player benefits the most from that - we're kind of on our own to assess how the psychological factors for both players affect the position.
The bold move is sweet but I am never making it on game 7 of 7 of WESPAC final against Eldar when we are 3-3
im hungry and sleepy mistaken this for a french food review video
Do the scrabble engines you're referring to use brute force computation or neural nets like alphago/alphazero?
They do Monte Carlo sims, but based on a static evaluation of moves and game state. No neural nets yet (except for one beta version of an engine that was never published). The first developer to thorougly apply machine learning to Scrabble will take our level of computer analysis to a whole new level.
From a little bit of research just now, it seems like existing Scrabble engines don't use neural network computing/ deep learning techniques. At least not yet.
I imagine that applying those techniques to a game without perfect information, especially one as complex as scrabble is very difficult.
@iankrasnow5383 Yeah, as I see it, the likeliest way deep learning can help scrabble is in quantifying board openness/closedness in any position, which could allow for improved Monte Carlo simulations. I see this as one of the three big things that Scrabble engines could improve upon significantly with today's resources. The other two big areas of improvement as I see it are (1) to have engines iteratively scan the search space (where plays can be mapped to a 3-d space based on points, value of the kept tiles, and board position quantified by the neural net), and the engine searches throughout the space and identifies the types of strategies that make most sense for the turn at hand and keeps zooming towards a smaller set of plays as more Monte Carlo iterations are performed, and (2) instead of performing, say, 10000 Monte Carlo iterations on 10000 permutations , perform 10000 iterations on 100 permutations (i.e., 100 permutations with 100 different strategies each time), and perform a weighted average on the 100 strategies within a permutation based on the fruitfulness of it
@@morrisgreenberg5223 I just don't know how this kind of AI works against human players with games that involve psychological plays, imperfect information, and occasionally bluffing, like this or especially games like poker. There's no one-size fits all "best" move against every player sometimes. Computer advantage mostly comes from the fact that they have perfect word knowledge and always find every single high scoring play, so only a few humans are at the level where they can compete on strategy alone.
The way I see it, they would need to train an AI using real human games rather than simulated games. And even then, it's hard to know how they'd pick up the right parameters to make an effective strategy. Overfittiing or underfitting seems likely in this kind of problem.
@iankrasnow5383 A few things:
(1) Poker AIs show mixed strategies as the best moves the majority of the time (i.e., 30% of the time do action A, 40% action B, etc.). So while they don't suggest "best" moves like in chess, they estimate optimal strategy still (and diverting far away from it tends to not bear fruitful).
(2) My last of the three big improvements in my mind gets at this, where if you do a deeper dive into strategic responses by an opponent (rather than assume a strategy based on highest equity responses), you have more ability to assess what types of plays are best at combating a range of reasonable strategies (where reasonableness can be approximated with reweighting the different strategies based on their performance)
(3) Yes, overtraining neural nets can have issues generally, but I think just doing it to estimate how conducive a board position is to scoring points isn't as problematic here (and even with computer-generated games, I don't think the bias will be large compared to human play with perfect word knowledge and play recognition). Any square on the board can be encoded as occupied, hookable (where similar to this PRETZEL video, we can encode this information in terms of number of hooks available and face value of what's being hooked), or unoccupied and unhookable. As the response, use the number of points that were scored in the rest of the game. Part of the reason why this is still fruitful in my mind is because a very high percentage of best computer plays are still the best human plays, and there isn't an obvious directional bias in terms of openness vs closedness across other plays