Scrabble GM vs. AI -- the Rematch! Game #35

[@AmaranthRBY]

LOOIE is such a great example that Scrabble is not just word knowledge! Basic word that everyone will learn within a couple months of starting (as are the other crossing words), but you gotta have that next level eye to actually find a play like that. Love it

[@axcertypo]

Nice game. According to Macondo LEADENs isn't really a mistake, but I agree that it should be. X(I) because the A *does* hit! You got close seeing CENTURION and REDUCTION. Maybe KINDA D4 instead of NAKED. Or fishing the K is probably fine. Those one-tile fishes sort of imply a lack of an S in my experience, so blocking that 7s lane has a lot of upside, despite giving back the lane hooking onto AM. HEAVY seems best as you discovered. FLA(K) was your biggest mistake. Those new 4s are tough. According to the engine WI(R)E is better than WO(R)E despite not setting the S up. The O just helps you tremendously on those worse draws. I don't think ODD(I)TY really works because of how few outplays you draw. According to Macondo ODDITY still only wins half as much as GUYOT even with an S inference. DO is interesting but I think ripe for getting completely outmaneuvered with 2 in the bag. Hard to say without looking at a bunch of positions manually afterward. And you made a one-point endgame mistake. Best sequence is cute, but not worth looking for when you know you've won

[@mackmeller]

Ah yeah forgot about RECAUTION! Interesting about WIRE being better, but yeah I could see it -- I did get pretty lucky to have a big S play there given my leave of SIRR. A good reminder than while I > O in a vacuum there are certain tiles that the I is far more toxic with than the O!

[@AlexDings]

Every normal player: Ugh, of course my opponent gets to play QUAD there Mack: *Should I fish for QUADRIVIA?*

[@Para20Site]

KZbin increasing the tension by starting ads exactly when Mack spots the LOOIE play, so I have to wait longer to see what he spotted.

[@G3N3515DM]

pssst get an ad blocker...

[@mackmeller]

Hahaha I promise I did not plan it that way

[@prodbymorii2058]

Mack, just curious - how many hours a week do you dedicate to studying words? Or do you only prep in advance of tournaments?

[@mackmeller]

These days, mostly I just prep in advance of Nationals -- in fact, before 2023 Nationals, I hadn't really formally studied in like 5 years (no exaggeration). When I was younger, like 11-15, I was studying much more, probably 2-3 hours a day on average as I was learning all the words. But once I learned all the words I found I didn't forget them very much, so I never did cardboxing or regular systematic reviews. For a period of about a month or two before nationals last summer and this summer, I'd say I studied 3 hours a day on average doing a review of the full dictionary before nationals, and also some specialty lists that are notoriously confusing such as -INGS and -LIKE words. I'm back now to not studying, and probably won't resume until next summer. Most of my Scrabble practice during the non-nationals parts of the year these days comes from my games for my KZbin channel and other practice games (sometimes against local players in person, or against the computer).

[@coolnath99]

great game. 🎉

[@mackmeller]

Thanks!

[@asdfasdf4924]

As are really valuable tiles. I would not have played NAGA; just NAG instead--there were only 4 As left!

[@Columbine-en3qc]

27:07 bot finding incredibly rare hooks in the shape of W(BRUCITE) 😂

[@TheGr1mmLord]

:)

[@davidg5506]

I really should have seen LOOIE because I was looking at just LOO as a decent option. Great find! fwiw, flak rhymes with lack :)

[@arcieplays9040]

Axcertypo has already done a nice analysis, but I did my own :) Exch NQW is correct. BestBot's 8H VIE is perhaps not, narrowly winning the regular sim but falling to 5th (55.50%) behind sim winner Exch EIIOV at 56.13% in the inference sim. LEADENs narrowly wims 0.05% ahead of LEADmEN. All best so far. 1K A(N)GLO sims 2nd at 53.94% behind J2 GLOAT at 54.22% (110k iterations), NAGA sims best, narrowly over NAG. MOTEL sims first by 2%. D11 NAKED sims 4th (57.56%), behind TAKIN (58.56%), KINDA (59.15%) and K(EtA) (64.40%). Big Win% difference (O)ESTRINS, (L)OOIE, and H(E)AVY all win their sims by a considerable margin. FLA(K) sims 3rd 3.75% behind LEW(K), clear mistake. Lastly, BestBot GUY(O)T wins both the inference sim, and the regular sim, however the inference sim strangely does not show a big increase in S frequency. Simming with an inference of S, has 2A GO win comfortably over GUY(O)T. 3B WI(R)E sims 3.5% ahead of 3B WO(R)E, despite the lack of S setup.

[@mackmeller]

Thanks! Yeah the most interesting thing to me is WIRE being better despite the lack of the S setup, goes to show I was pretty lucky to hit something as good as SWIPER with SIRR and that the I can be so much more toxic than the O even though it's better in a vacuum

[@amitaistern1881]

24:23 with an r draw the bot gets crustier through the blank s

[@su1tcase622]

4:20 Why can you not play (s)ago? I'm extremely new to competitive Scrabble as a whole and would love some insight

[@100beep5]

Probably could, but there's no good reason to

[@snowman99tetris]

The l isn't worth 14 points on leave lol

[@Goryus]

When Mack says he doesn't see a place to play it, he means he doesn't see one that is worth considering. There are also plays like AGON but they all score 4-5 points. An average score for a scrabble pro should be 20-30 points.

[@mackmeller]

I can, but as pointed out below, I'm giving up way too many points (AELT as a leave is a little better than AET, but not 14 points better). SAGO is also very risky putting the O in the TLS-DWS like and also exposing the SAGOS hook.