Scrabble GM vs. AI -- the Rematch! Game #26 

at this point, mack just needs to beat the white sox' record through 100 games this season (27-73)

one thing that's really stuck with me as I've improved at Scrabble over the past few years is that every heuristic has an exception. I think OR really exemplifies that - opening a huge bingo lane while up 30 with a generally-ok leave is not something that most top experts would do. Looking at that position, I don't think I would have had the guts to make that play either. But this is the great part about computer analysis tools: it shows us points where our intuition fails. There are definitely reasons why BestBot would overvalue OR, given how its algorithms work, but I think it is still a very interesting position to learn from.

Definitions of Interesting Words Played in Game 26: JOLE (22 pts) [noun] - variant spelling of "jowl"; cheek, the boneless cheek meat of a hog [from Old English] FEZ (34 pts) [noun] - a brimless cone-shaped hat that has a flat crown usually with a long tassel attached, is usually made of red felt, is worn by men in eastern Mediterranean countries (as Turkey), and has been adapted for women's hats in Europe and America [French, from a Moroccan geographical name] DEIFICAL (76 pts) [noun] - variant of "deific"; divine, godlike MANTLING (74 pts) [noun] - a heraldic representation of a mantle behind and around a coat of arms [made up of French and English combining forms] CASERNS (84 pts) [noun] - plural of "casern"; a military barracks in a garrison town [French, from Provençal (a Romance language of Southern France, ultimately from Latin] UNLAY (35 pts) [transitive verb] - to untwist the strands of (as a rope) [made up of originally English parts] WHANG (32 pts) [intransitive verb] - to make or produce a resonant noise [alteration of another English word] DITZ (42 pts) [noun] - a dizzy person [alteration of an English word of unknown origin] DRONGOES (86 pts) [noun] - a plural of "drongo"; a bird of the family Dicruridae native to Asia, Africa, and Australia [from Malagasy, an Austronesian language of Madagascar] BOITE (40 pts) [noun] - variant spelling of "boîte"; a nightclub [from French] KAE (21 pts) [noun] - (chiefly Scottish) jackdaw [from Middle English (northern dialect)] BENE (24 pts) [noun] - variant spelling of "benne"; another term for sesame [perhaps of West African origin] LIRI (6 pts) [noun] - plural of "lira"; the basic monetary unit of Italy until 2002 [from Italian]

You play OR to distract from bingo defense given the vowel-heavy pool. I play OR because I don't know that TROGON or DRONGO are words. We are not the same.

I'd like to think I'm capable of playing OR. I do think it's definitely the best play here. Against a lot of humans I'd have to bail out with TROGON or something, because the risk is very high and I'd feel so dumb when this backfired, but it certainly looks like the best play to me, honestly by quite a lot. For me, one of the reasons I avoid playing as open as other top players is because I've always felt like this sort of aggression is absolutely optimal once you're on these sorts of boards a lot of the time, and playing stuff like OR does make me a little sick to my stomach, but once here I think you kinda have to do it. The pool smashes the *** out of your range, you're not ahead by enough to outlast bad draws or opponent good draws, and you can always fish again if you miss.

One thing about OR is that it keeps five different tiles, which is a nice insurance against the pool (full of duplicated vowels, plus Q and V). It risks the counterbingo from you but the upside is that in ensures the bot is fine is almost all other scenarios. I like making those kind of plays (although whether they are sound when I do them is another question 😉)

So used to content creators say they 'privated' videos instead of saying they set the videos to private that I was near certain it was a word. I guess there are always lessons in these videos regardless the outcome.

Spoiler cinderblock. I love me some good ol fashioned concrete

Never seen Mack straight up stop talking to react to a tile draw lol. Crazy that the 5 Es are a sidenote of this game because OR happened

Being allowed to play phonies would make,the bot even more formidable and better simulate an over the board game

After OR, my immediate instinct was playing PA(C)T given the range. If you assume after OR that the Bot will bingo on either the top or the bottom more than 50% of the time, then VIPER is basically an insta-loss that often. If you assume the Bot's range is less good to the point where its optimal strategy is to open a second lane to long-term give itself options (as it seems like the OR play is really to minimize the scenarios where you have a strong enough range to bingo from the D and it is drawing dead to your newfound lead), then opening a third lane could flip the concept towards your advantage in terms of tempo. PA(C)T makes the rest of the game a lot more random which seems advantageous in a situation where you're down 30 with the opponent either having a great range or needing the volatility to outrun high-point bingos in the short term.

5 E's, lol, actually hilarious. And your reaction, haha

Do you know how often the bot is updated? Is it possible it's just gotten that much better since the last series?

Spoiler block 🧱

what a fun episode!

I like OR a lot. Can’t say I would always think of it over the board or have the cajones to pull the trigger, but it’s great. While BB has a lead, it’s still pretty precarious after something like TROGON or DRONGO with lots of potential bad draws, bingoes you could hit, X bomb, etc. Turning off the jets there seems premature, OR seals up the game immediately a solid 40% of the time

“PAW, MAW, it’s close” You’ve just summoned the furries, Mack!

I would not play OR. I get the idea of distracting, and opening up two lanes on a board that has only one, but it seems too aggressive to me. I'm not quite a top player, though. On the UNLAY turn, I liked 9C AYIN for 28, setting up the L front hook, and also setting up the possibility of LUT(Z) with a T draw. Also, UNLAY seems to close off the bottom on a board that is already dying, as there are no Cs or blanks left for MYC. It is quite a big equity sacrifice, but I like the board after AYIN much more than after UNLAY at a ~30pt deficit.

ENTERTAINING TANGENT MACK

Nice game. I think OR was a distraction.

Was "or" a godlike mind game?

ENDGAME SPOILER Wouldn't the chance of drawing 5 E's in the endgame be much rarer than 15 choose 5 (1/3003), since your opponent is more likely to have E's since they are a good tile, thus making getting an E rarer?

UU = -18.8 WU = 13.0 It sounded like you had some kind of noise in the background there, it would be nice if you could solve that for next one

i think the bot played OR just to flex 💪

Is teenie valid?

The bot draws four Ss and both blanks while Mack draws 5 Es on the same rack. (Another "I'd like to sell a vowel" game!) And then the bot comes up with a risky play of OR - because it knows Mack's rack and that he can't exploit it? (The bot is in control of tile draws, so it knows the entire bag, and therefore what's on Mack's rack.)

Hello! Tough loss, I've analyzed both you and BestBot's moves with macondo (ply 5) and here is my analysis!: POUC(H) sims 5th at 39.61% behind POW (40.27%), CAMPO (40.64%), MAW (40.7%), and MOW (41.13%). BestBot's J(O)LE sims 2nd at 66.75% behind J(O)LLIED at 67.32%. WAVY is best, and BestBot's 5I ODD sims 4th (59.25%) behind 9G OLD (59.3%), 5I ODIC (59.38%), and 9G ODD (59.55%). FEZ, (D)eiFICAL, MANT(L)ING, cASERNS, UNLAY, WHA(N)G are all clear best moves. Exhange EEEEI is preferred (17.37% vs 17.32%) over exchanging EEEE. BestBot's ROM sims 2nd at 87.4% behind MORN at 87.46%, and DIT(Z) is best. 14E OR sims 2nd at 77.02% behind 3H (T)ROGON at 77.37%. Still, a very interesting play totally payed off for the bot. VIPER sims best, even using macondo's inference feature. Simming with perfect inference, plays that block the bingo rows up top sim only fractionally better than VIPER. After (D)RONGOES, all other plays before the endgame were best, and all chances for Mack slipped away. All the best in the next game, Mack.