Ultimate Scrabble battle: Grandmaster vs. AI! Game #48
HTML-код
- Опубликовано: 19 окт 2024
- In this series, I'll be playing the AI-powered BestBot in 100 games of Scrabble! Who will prevail? Watch the videos to find out -- and hopefully pick up a few tips too!
BestBot is the upcoming ultimate Scrabble AI from Woogles.io, to be launched in 2024. For questions, please email woogles@woogles.io.
Want personalized help taking your game to the next level or a fun gift for a friend? Check out www.mackmeller...! for more info or email me at mackmeller@gmail.com!
![ARGENTINA vs. BOLIVIA [6-0] | RESUMEN | ELIMINATORIAS SUDAMERICANAS | FECHA 10](http://i.ytimg.com/vi/C3wPtKNbAOA/mqdefault.jpg)
This comment is used to prevent spoilers
And this comment is used to reply pointlessly to a comment used to prevent spoilers
Fun to see GENTILE followed immediately by ERUV
I think the bop play is fine. Mack didn't play diktats himself, so increased chances he doesn't have the s. Also, even if Mack does, he may not have a great rack to double double, whereas the bots ability to double double seems much likelier. I think the play is about neutral on an even score, so definitely good when behind.
Yeah, good point -- I think I didn't consider seriously enough the fact that BOPS plays that aren't bingos really aren't very painful, and bingos aren't all that likely
Yeah the BOP turn is very interesting. I don’t think I’d have the guts to play that, but I can see it being worthwhile. I did differ a bit on my inference of the bot’s rack though - I don’t think assuming multiple S’s is correct. Yes, it having more of them means it thinks you would be less likely to have one, so from its perspective BOP might be safer. However, if it had multiple, then the downside for its leave of putting the S down is a lot slimmer than if it only had one, since keeping a duplicate S likely wouldnt be worth sacrificing 9 points from not tacking it on. With only 1 S, the value of holding it can reasonably be a lot more than 9 points, which is backed up by static leave values and probably enhanced given how valuable the S is on that specific board. I might have inferred a reasonable likelihood of a blank or two as well, since that would make it less likely for you to hit the hook without the thoughts of duplication, though in this case it still seems to think it is worth it without either blank.
That's a great point about having multiple Ss making playing BOPS immediately a lot better on equity, definitely didn't consider that! And yeah could definitely see a blank as well in a lot of cases
I don't know how Quackle works, but it surprises me that it would get an endgame analysis like this one wrong. How is that possible? I can understand how it might miscalculate very complicated endgames, but this one isn't excessively complicated, is it?
Fair point -- I honestly don't know the full technicalities of how it's endgame solver works, but I have seen it mess up occasionally with 1 in the bag in situations like this. When you play out each individual endgame based on each draw it plays them out correctly, just doesn't evaluate the win % right with 1 in the bag. Definitely strange though
I think it’s maybe sometimes reasonable to guess the bot has a blank if it’s taking really long since it drastically expands the number of possible plays it could make/routes it needs to expand, even if it’s gonna quickly find all the words
Commented that before the post hoc analysis so i don’t actually know yet if it had the blank
Something to consider with BOP is that the bot's algorithm probably overrates its own scoring potential should it actually hit a bingo next turn. Probably the bot is happy to set up BOP because it sees that its own score increases by a considerable amount on average the following turn, but it doesn't factor in that humans will play stuff like ARF to block it way more often than not.
Then again, maybe the bot is right not to worry too much about these kinds of plays - if it's sacrificing 4 points and you sacrifice >4 points' worth of equity to block, the bot comes out ahead. So probably the scenario where BOP is immediately blocked is kind of moot (he sacrifices a bit to set up, you sacrifice a bit to block, game continues); the upside is purely in the scenarios where it stays open, and the bot is definitely more likely to hit the spot than you are since it's already holding the S.
All in all I can understand how it could be good for win%. Especially to guarantee another S spot should DIKTATS get blocked
Would love to see quackle's take on AYE vs YOU vs AGIO, that one was really interesting to me. Intuit agrees with AYE, and I was feeling good about OUZO as well
Thanks! Yeah great point about me sacrificing equity to block, it's certainly something I'm going to go considerably out of my way to block often at the expense of a bunch of equity (as I indeed did). Even though he really needs a bingo to score big with BOPS a bingo there is so many points I kind of have to block given his range is so bingo-heavy.
Yeah AYE feels kind of forced, as bad as the leave is I really need points there
It would be interesting to pit BestBot against HastyBot in 100 games, since they run on very different algorithms.
Any experienced human with decent word knowledge (or using a dictionary to level the playing field) should easily be able to learn and exploit Hasty's weaknesses. But would this bot be able to do the same, or would its unconventional plays (like the BOP turn) be punished by Hasty's purely equity- or point-driven algorithm?
Love these games, very educational. I don't think I would ever have played BOP there...
Given that there's already an S hook on the board, isn't it actually safer for the bot to open a second one even if it thought you had an S?
Funny to have one each of AEIOU and Y!
I would play BOP but not in a game where I’m a significant favorite. I want to make equal/better players question WTF I’m doing, feel like they have to block, and sometimes cash in myself. And yeah sometimes I’ll get hit hard by it. But like my grandma always said, the next best thing in life to gambling and winning is gambling and losing.
Yeah, I would have played it as well, especially against another human at the casual level, because they might even burn a blank just to block it, and I would still have other parts of the board to work with.
I would also make similar, forcing plays when I'm fishing for funny words and need to divert my opponent's attention from certain letters on the board. But again, I'm just playing for fun, and I'm someone who would keep OUVV to fish for things like UVAROVITE through AR...
Yeah, it certainly will cause a huge number of paranoid blocks (of which ARF was kind of an example), and I guess also barring bingos, BOPS plays really don't hurt all that much. The more I think about it the more I agree with it
@@manaphylv100 I played Uv(AR)OvITE with 2 blank Vs as an only playable bingo once!
@@mackmeller I definitely would not have considered that word without at least one real V in hand!
I often have trouble with 2 blanks when I don't already have a word in mind...
The bot was quite obviously buying time on the BOP turn. He kept a good hand, but not that good, so he made a forcing move to keep the top and right open. He was forcing you to spend your S, and if you didn't have an S, you would likely open something else for him.
I would definitely have made the same move as well if I had the bot's hand, especially since at the casual level, your opponent might even burn a blank to block it.
There's a similar concept in most chess games that involves sacrificing material for space/position and time. (A famous opening in Chinese chess sacs a horse/knight for rapid development.)
Makes sense -- I definitely wouldn't burn a blank to block it, but I did indeed end up sacrificing a considerable amount of equity, so the plan did pay off
I wouldn't either, unless I had both blanks and multiple high-scoring tiles that don't go well together. But at the casual level, people are quite likely to burn a blank to block your setup, or even just to take a 3xW.
I have an idea that could be used for quackle. What about playing a game until you can't play any more. No more valid moves. Each draw is equally likely as a starting bag? The board would look CRAZY by the end of the game.
Hmm, not sure I exactly follow -- do you mean a bag with like multiple sets of the usual tile distribution so there would be way more tiles than spaces on the board?
I think the idea is to have a sort of "magic" bag that replenishes whatever tiles you draw from it, keeping its contents exactly the same throughout the game. This could maybe be approximated in Quackle by just multiplying the bag contents many times over (hundreds or more), so that even as you draw tiles the distribution stays approximately the same.
I think the rationale behind BOP in that specific spot instead of to the left of MIDCAP is to keep the G of GAB open as the bot is holding IN. Also, the positioning of DIKTAT doesn't allow for a bingo ending in S to fit, but if your assumption that the bot was holding more than one S was correct then any bingo would have fit, thus opening that second S spot wouldn't have been necessary for the bot.
As for the endgame, I would have assumed after LITAI that the bot was holding at the very least the second A, because I don't think it would have "wasted" it and forgoing the ZA hook. I was already worried about plays like JADE on the AYE turn, so if anything, OUZO was technically a 12.5% play rather than 25% because of that duplicate A assumption. Would DIE or REZ (or similar plays) on the I or Z of ZINCS respectively have been solutions to stop the bot from playing there altogether or was JOUAL simply too good of an alternative to outrun you anyway?
Good question -- the problem is DIE or REZ do block effectively but at way too great an expense to my rack. I'm going to have 2 Us left and all kinds of other garbage and have no chance of going out next turn, so the bot is going to get 2 more turns and should be able to out-in-two quite easily and win. In these close endgames it very often comes down to who goes out first, since generally it's rare for 1 play to score more than 2 plays plus the unplayed tile penalty, so I need to play something like OUZO that gets rid of more of my vowels and gives me a chance to go out next turn .
16:52 euoi next to hornists?
Collins only!
@@mackmeller oh, okay. I have never played nwl.
Against a human, is there any scenario where sacrificing the bingo and playing SENTINELLING is correct, because it has reasonable odds of eliciting a challenge and earning you an extra turn? Perhaps even more so if you had history against this particular opponent, and had reason to think they're more likely than the average opponent to challenge in this scenario? The word looks so weird, and given the score, it's not unreasonable you'd be grasping. With all that said, I think the answer is still no, even though I want it to be yes b/c getting a 12 down is so tempting. Worth thinking about though, at least. If there was 1 more N in the bag, setting up a potential -ING lane, perhaps it would be slightly more worth it.
I’ve learned that in almost all cases, the bot plays a word in the pre-end game that leaves one in the bag.
Bop is fine. You should play the odds. Bot will get that hook more than you will.
If I lead by a bit then no I wont do it, but in a close game when my opponent is unlikely to have S? Sure. Yeah it feels bad to get hit by worse case scenario, but I never hesitate here.
Spall Year Blah Cur Calm Meant.
Spoiler blocker comment