Scrabble GM vs. AI -- the Rematch! Game #7
Vložit
- čas přidán 9. 06. 2024
- The Scrabble AI BestBot got the best of me in my 100-game Human vs. AI Ultimate Scrabble Battle, but I'm not ready to cede to our AI overlords! Introducing... the GM vs. AI rematch!
This 100-game series, running every Monday and Wednesday at 5pm ET for 50 weeks, will feature 20-minute games against BestBot with post-game analysis. Hope you guys enjoy, and wish me luck!
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.com/lessons! for more info or email me at mackmeller@gmail.com! - Hry
Notable Words Played in Game 7:
MOTILES [75 pts] (plural noun) - in the field of psychology, individuals whose prevailing mental imagery is motor rather than visual or auditory and takes the form of inner feelings of action (such as incipient pronunciation of words, muscular movements)
ABOITEAU [74 pts] (noun) - variant spelling of "aboideau," a type of tide gate or dam used in Canadian regions to prevent the overflow of water into marshland [from Acadian French]
SHOTE [48 pts] (noun) - variant spelling of "shoat," a young hog of either sex, especially one that is less than a year old [Middle English, of uncertain origin]
ZINGER [52 pts] (noun) - (chiefly American, informal) a pointed witty remark or retort
SERJEANTY [110 pts] (noun) - variant spelling of "sergeanty," any of numerous feudal services of a somewhat personal or menial nature by which an estate is held of the king or other lord distinct from military tenure though it might involve service in war and from socage tenure and varying greatly with different holdings (such as steward, marshal, constable, chamberlain, or esquire)
GREENWAY [92 pts] (noun) - a corridor of undeveloped land preserved for recreational use or for environmental protection
Awesome, thanks for all the info!
finding SERJEANTY in 3 seconds flat is insane
Haha thanks! Been studying a lot of 9s lately :)
Finding the 9 of "serjeanty" so quickly is incredible!
Thanks! Been studying lots of 9s lately 😀
sorry about the slowroll. I didn't fix that bug yet because I actually can't think of the best way to fix it, lol. I can tell you what the bot is doing - it's trying to solve the endgame after 2 plies, 3 plies, 4 plies, 5 plies, etc ("iterative deepening") until it gets to 14 plies. It sees that GREENWAY is by far its best play, but it needs to verify this at every ply, and it's always trying to calculate out all possible endgames (i.e. if it slow plays, with the blank, it has a gazillion possible plays). I do have a pruning algorithm that I've been tuning up for years but when you go that deep (14+ plies) it's always going to be slow if you have many thousands of possible plays. And yeah, I think it can't actually stick you with the V, so it should only be going down 6 plies at most (you can also slow-play) but it keeps calculating the same thing over and over again.
One shortcut I have in mind is to have it return the best play it has found so far if it keeps finding the same play maybe 3-4 times in a row while doing the iterative deepening. That should work most of the time. Another easy shortcut is to just take the sure win when it finds it. It will sometimes just blow some spread but it might be cheaper to do it this way in terms of computation time 😅
Haha all good man, and thanks for clarifying! Just a fair warning (without spoiling the details) there's a far more epic slowroll coming up in a few games and I did maybe get a little grumpy, but don't take it personally :)
the scorecard doesn’t accurately display how masterfully you’ve been playing these games!
Is there any way to get BestBot as the analyzer for your moves? (Maybe export the game to another software?)
I would really like this!!
+++
Macondo (=BestBot) is free to use (it's also open source). It only runs in the command line but you quickly get the hang of it. In my own series vs BestBot (in German) it's been a regular feature that I'll analyze the game with the engine afterwards, and viewers seem to appreciate it. To be fair, Mack is an incredibly sharp analyzer himself and I'm not at the same level, but I do think it would still improve the videos
Unfortunately (as Alex noted) not right now directly via the Woogles screen, but hopefully soon!
Dude, I give you credit for deciding.
Any chance, you can play bots that always play the highest scoring move or play the most tiles? If you can find them.
What a play!! Nigel would have been proud. Or maybe not because I'm not sure he's proud of anything, but still.
Hahaha we may never know... thanks though!
that was an exciting finish. at least you get style points for serjeanty.
Thanks!
With the vowel/consonent ratio 24v/41c, that is very consonent heavy; I would 100% keep IOSP and exchange UUO on that move
Just exchange IOOPUU and keep S is fine. 24v:41c is not a super consonant heavy bag
I would have played DUDE to fish for AR (CINERARY), AV (VICENARY), or PL (PRINCELY), since I would consider that to be my best chance, assuming the bot had the blank.
AL, AT, EF, ET, and ID could also hit the T with CLARINET, INTERACT, REINFECT, RETICENT, and INDIRECT.
I'm sure I missed quite a few, so the chance of hitting something wasn't too bad, but some of those to the T might lose as it would give too much counter-play.
Interesting -- I'm not sure I'd assume the bot had the blank after CRIB, though, it could easily have just had a consonant-heavy rack without the blank. And I think DECIDING should win fairly often when I draw the blank, which is much more often than after DUDE since I play 6 tiles instead of 2. There's also the factor that the bot may sacrifice to block the bingo lines if it has the blank, figuring it'll be able to outmaneuver me in the endgame assuming I don't bingo. Lots to consider here for sure!
@@mackmeller Oh, I simply assumed that the bot had the blank as the worst-case scenario. In other words, I wasn't counting on drawing it to have any chance at winning.
But yeah, good point: he might indeed sacrifice to block. I forgot this is BestBot, not Hasty, and he seems to be smart enough to see through fishing and setup moves, especially in the endgame.
Great effort! I think S(ER)JEANTY is one of the best 9s I've seen a human spot in a live game.
Thanks! Yeah I went from "dangit I didn't want the J" to "holy s*** I have a crazy 9" very quickly there hahaha
I hope you got kids already Mack. Because you’re getting bagged so much in this series that you might never have them again.
I was surprised you weren't considering privy, but then you found triple which was better. Sweet game.
Thanks! Yeah I think PRIVY loses by a couple even if I draw the blank since I don't score as much with my outplay with DLT? as I do with DV? after TRIPLE -- given what was in the bag I was dead either way though.
Unrelated to the great video, but I tried to make a rectangular board in Quackle, but all the bonus squares kept disappearing except from the ones inside the top-left largest square. Square-centric Quackle?
I've had the same issue, unfortunately, which is why my really cool rectangular board modified Scrabble variant ideas haven't materialized :(. Must be a bug with Quackle
Bot's final rack would allow it to play WANDERER... Actually poking fun at you?
Or paying homage to the great Will Anderson, aka wanderer15?
@@mackmeller I'm not sure, because you didn't seem incredibly... happy... about the result of this game 😛
There certainly are some weird OUT- words. I recently missed a bingo of OUTSTRIDED. As if OUTSTRODE wouldn't be weird enough...
Wow, I totally would've challenged OUTSTRIDED, especially since STRIDED* is phony!
Amarone is pronounced a-muh-row-nee. A wine made in the Veneto region of Italy.
It was Hannibal Lecter's preferred accompaniment to (human) liver but they changed it to Chianti in the movie.
Ah good to know!
Well this was an exciting game! I analyzed it with Macondo and here are my main takeaways:
G6 vs I4 MOTILES is close in sim; but G6 MOTILES performs better when adding inference, disagreeing with the bot.
EMANAtOR is best and WAG(E) sims 2nd behind 3L PAW by a mere 0.2%, definitely fine.
On 9 Plies 3K EAU outperforms exch AEIIU from the bot by .5% another disagreement between them.
PIOU(S) narrowly sims best, 7K KOA sims 6th, 1.15% behind 10F KOU(S)SO and 1% behind H10 OAK.
BestBot's 11D GYV(E) sims 3rd, 0.6% behind 11E VE(E)R/11C VERG(E)... Perhaps these plays are better at avoiding comeback scenarios like SERJEANTY
Macondo is against SHOTE by but... I mean... how else does it expect you to make a comeback?!
S(ER)JEANTY is a great find! Puts your winning chances back to 36%!
Possibly the largest mistake this game, Macondo has N10 CREDI(T) simming 10% higher than 3A DECID(I)N(G)! Although to be fair, by that point the bot had the blank so it likely doesn't matter
Now time to watch Game 8!!!
holy crap SERJEANTY
I don't suppose it's possible but I would be curious if best bot with your draws would win vs best bot with best bot's draws, your draws seem to be quite hard in a number of those games making the scoreline seem very deceptive. But surely you'll just draw better next game, right!?
Yeah probably no way to know, but there are still 93 games left for me to draw better! :)
He slowrolled again...
First like
Ooh! It is posted at 2:00 PM
Yep same time every day!
@@mackmeller I got out of school early! That’s why I noticed. I got out 3 hours early
29:17 “doesn’t put an R in the 3x3 lane” well, not the triple word, it does the triple letter