Scrabble GM vs. AI -- the Rematch! Game #3
Vložit
- čas přidán 26. 05. 2024
- *Prediction contest is open until 5pm ET, Monday June 3rd - enter here: tinyurl.com/3tt39dpp
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
Fun spoiler fact of the day:
Out of the 20 named species of armadillo, 19 of them have been found to live in South America. The familiar nine-banded armadillo is the only native armadillo in the contiguous US.
This one I definitely did not know! Very familiar with nine-banded armadillos but yet to encounter any of the other 19
Unfortunately this spoiler blocker did not work, as the comment about the triple triple showed up instead
@@almightyhydrasame
@@almightyhydraCZcams is stupid sometimes.
0:42 Holy nga! No f-ing way! Well, we can add Nick Punto to the list of Major League Baseball players whose first and last names are both also real words.
27:00 If you didn't block with WOLF the bot would've gotten a triple-triple with pRO(F)iLER
Ah good call, didn't think of that one!
First!
The decision about where to play MORAINE/ROMAINE was really instructive on how to play when you're losing, super impressive even if it didn't end up mattering. I have no idea what's right but i feel like C1 MORAINE has the best upside.
Thanks for watching! Yeah I'm still fine with my call ot play C1 MORAINE even though it backfired terribly haha
Tough game at the end but liked the aggressive decision
Thanks! Yeah I don't regret it even though it blew up spectacularly haha
Nerdiest was an amazing quick find for you after the bot played dementis.
10:06 The only explanation I can think of is maybe the bot was looking at the bottom left TWS with a Z overlapping the E (as the Z is still alive in the bag). In which case, the D takes a different set of vowels in front of it (A,E,I,O) than the R (A,E,O,U), but there are no 3 letter words starting with "ZU" whereas ZIT, ZIN, ZIP, ZIG are all valid. Aside from the Z maybe it was looking at all the other heavy tiles that could go in front of the E (at this point, H, Y, W) and these all seem to have more 3 letter words with "I" than with "U" as well (after all, U is easily the worst vowel). That, and there is only one U left in the bag anyway but there are 6 Is left.
However, I'm not sure if the math ultimately makes sense in terms of win % when we compare it with the increased % of a hook with a bingo starting with S which would also hit the TWS at the bottom.
Absolutely sickening to spend so much time and energy on a turn and then the bot instantly slams down a rack that dunks on you no matter what lol
I thought you played really well throughout honestly, just unlucky.
Regarding REROLLER vs REROLLED, it's probably just glitchy, but I wonder if it's also considering plays on the A column that play with ID but don't play with IR*, stuff like WIZ for example is quite a bomb. It feels very tenuous to justify REROLLER but it's an upside at least... Bingos starting in S are not rare, but probably not common with both blanks and one S already out, so the threat of instant punish through REROLLERS is quite low, and the hook is easy to block in the next few turns (not to mention giant upside if the bot draws an S itself). Or in other words: REROLLER opens up the biggest possible punish, but REROLLED is a tiny bit worse against the 'average' punish in a consistent enough way to possibly make it not best... maybe?
I dunno. I'm really trying to come up with something that isn't just "the bot broke". Another consideration is maybe floating the D vs the R for 8s, but it doesn't feel like that would make a huge difference
Interesting! Definitely a more plausible explanation than anything I was able to come up with haha
i think it's very plausible that with 2 blanks the bot simply had too many possible plays/too many good plays to simulate and didn't converge on a clear winner between reroller and rerolled. idk how the sample size is determined but i imagine the difference is fairly small, it's just obvious to us humans since the words are so similar
@@thomascorey7284 The bot didn't actually have that many options, only a few permutations of REROLLE[RD] in that spot, EXPLORER through the X for one point less, and many much lower scoring options like LO(I)TERER through the I in JUICE, or any 7 playing down from B2 to B7. I don't know if the computing time would've been too bad, especially because the bot only took 14 seconds to make the play - it certainly had time to churn through more options if needed.
Like I said; "bot broke" is still the simplest answer and the most likely. But I think there's a few non-outrageous possibilities other than that one, and I personally don't believe REROLLER to be such a clearly worse play as it would appear to the human eye
@@AmaranthRBY yeah i don’t know the algorithm very well obviously, but i think the bot has to explore most plays even that score decently less, and having 2 blanks just multiplies that number by a pretty big factor.
Additionally, I think that unless the bot has a heuristic for "how many hooks does this word take" or something like that, it's very possible that a game-altering S hook wouldn't come up even once in like 20 or 30 sims (since this would require mack to get the S first which is not even particularly likely, and for the play to score well basically next turn). I think my point about how it's really obviously different to humans because the words are essentially the same other than the S hook might explain part of why we might be surprised by it's decision... not necessarily bot broke, but that humans are prone seeing to these kind of super obvious but really small optimizations. but i also think your points about how there are all kinds of other factors than the S hook are valid
(JUICE)RY opens up a sneaky M front hook and an S back hook making THRONGS
Think I prefer MORAINE just to avoid setting up easy scoring with e.g. DI(R)K. Obviously not scared of triple-triples but even with 2 bingoes we don't want the bot to outrun us
I laughed out loud after that triple-triple, classic BestBot
I was in the middle of a work discussion when I noticed this video dropped, and I had to abruptly interrupt my boss and tell him that something important just came up and that it required my immediate attention. I mean, I didn't really need that job... But, for now, my future looks like paradise -- at least for the next ~34 minutes!
Hahahaha so glad to find that someone clicks on Mack’s video with the same sense of urgency as I do
Hahaha thanks so much for your kind words, I really appreciate your support! This reminds me of when Matthew Tunnicliffe's Aerolith profile said something like "if I only got 32% on the sevens it's because my boss walked in" 😂
It kills me when you burn your own timer pontificating about bot mistakes when you could just do it in the recap, lol
How do you memorize the words that are in Collin’s and not North American. And why are they different?
5:49 True words to live by
Haha Scrabble can really be like life sometimes, part of why I think it's such a great game!
Can you reopen the form, i didnt realize i couldnt edit it 😅
The form should still be open, but I don't want to allow editing in general since I'm using timestamp as a tiebreaker and that could mess with it (i.e. the idea is that if you guess earlier you have less information, but you could get rewarded by having a better tiebreak). If you made an obvious typo though I'm happy to fix it, if so can you email me the details separately? Thanks!
I have no idea why you'd set the bot up for a triple-triple instead of getting the few extra points by playing ROMAINE toward the bottom. I don't understand the logic of that decision. I always play this game as thinking my play will always be the setup for the next play. Giving anyone a setup for a triple-triple is something I don't understand.
He explained it a bit but it is correct. After playing MORAINE at the bottom all BestBot has to do is block the last bingo lane.
I'm a fish but I really hated that MORAINE play. Even if the bot had bingoed there itself you would have then had a chance to score heavily along the top row. You would have only been 21 points down after normal play which doesn't seem enough to justify taking such a huge risk.
Matter of opinion and how you feel at the time I suppose but for me Mack made the right call. If Mack played at the bottom then BestBot would almost always score decently at the bottom right, blocking all bingo lanes in the process. And then block top left next go and would edge the win 90% of the time I felt.
Prediction MM 44 bot 56. Spread 1500.
Make sure to submit the Google form for it to count!
How long until you finally get suspicious? The bot has the absolute perfect rack every time. It even had TWO 3x3s ready depending on which bingo you played!! But I’m sure it’s totally random even though the same thing keeps happening after 103 games.
So you think there’s a supercomputer inside woogles code knowing what you are going to play and giving tiles to help the bot win vs that exact play
@@ccg-chatswoodcardgames9212 While OP's suspicion is likely misplaced, it would be very easy to make the bot lucky without fancy heuristics. Just simulate however many random racks you like and pick the best one via whatever metric the bot is already using to evaluate positions.
Bestbot always starts well in a series I've noticed. He's undoubtedly thinking, "this is a 100 game series, I must try my best to get a lead". Then after a few games he'll take more risks and give Mack chances to catch up a bit.
It is fair. Bestbot is simply stronger than Mack.
When the coder revealed the tile drawing code to show that it doesn't cheat, the code showed that the bot does have access to the entire bag. So while the draws may be random, the bot can infer the opponent's rack... and no doubt Mack could play at a higher level if he knew every tile on his opponents' racks.
I have no idea why the woogles server isn't programmed to give the bot only the info it should have: time remaining, tiles on its rack, the board, and the score.
You need to hold off on the series until the woogles server is redesigned such that it controls the tile draws and sends the bot only the tiles it draws, rather than allowing the bot to see the whole bag and draw tiles for itself. (The code that was shown does draw tiles fairly, but the bot having knowledge of the entire bag is not fair.)
With good tracking, the player also has knowledge of the pool of tiles unseen to them.
@@justin_tang Knowledge of the bag tells you exactly what your opponent’s rack is. Knowledge of the pool does not, unless the bag is empty.
It sure feels like that’s not the only unfair thing about the bot.
The bot doesn’t have knowledge of the entire bag
To all the players complaining about the bot cheating: chill. I've played against BestBot plenty, and the draws are definitely fair. BestBot is really good: it's at a level where Mack by his own admission isn't favored to win the series, where I'm also not favored in a series, and I'm not sure any human is favored against it in this lexicon at this point in time, though it's far from unbeatable and does make its own share of mistakes, especially in the preendgame. There are times when I find that I am learning from BestBot, and it makes better strategic plays than I would have post-analysis.
Mack is a great player, and he does a fantastic job of justifying his plays and decision making, but being able to justify your plays and decisions eloquently does not necessarily make those plays and decisions optimal. IMO BestBot is outplaying Mack, but that's nothing to be ashamed of, just like Magnus shouldn't be ashamed of being outplayed by Stockfish.