So which pro is limping K5o UTG?
No wonder Pluribus is winning.

Snowie raises LJ/HJ/CO at 17/20/27% at the lowest rake effective

they give a 50k freeroll to incentivize good play, but if a player is down money it makes sense to take riskier, higher variance lines to try and make a profit.

it would make sense to start everyone with $x in chips. whatever you have left you can keep, this would ensure good play for the entire time. if it is this way already then sorry for the waste of time. they didnt make that clear or i missed it.

i guess the only way to prove this bot works is to set it loose anonymously in a high stakes poker game. that would prove bots are capable of beating humans. but im pretty sure that was proven already. maybe not vs linus and otb 3 handed but in 6max games with a whale sure.

Why the discrepancy between the paper’s stated bots winrate and the winrate shown in Pokertracker when importing the hands?

Really appreciate this.

Already done with Oborra

Thanks, I'm really glad people are getting use out of it!

lol thanks, there's some good content on there, although I should try to speed up the bi-yearly post cadence.

You probably already figured out that the player with the name "Pluribus" is Pluribus, and the rest are humans.

Haha, sorry to scoop you on this! I was really gunning it down the stretch to be the first one to get it in

Ah, good. Glad to get confirmation that that wasn't a big bug on my end.

LOL - alright, in retrospect that was very fishy. I'm glad someone took a chance and downloaded it and saw that it was just a bunch of text files. Also I should have mentioned that you can see all the game logs in your browser here.

Oh, and if anyone ends up using these hand histories or the program to make something, it would be really cool if you could acknowledge me or link to the project page, because it would bring me a lot of joy.

Appreciate it lots, thanks
Can anyone manage to import the hands into hem2? I keep getting errors, even after separating the HH's when they're bunched up.

This bot isn't as much of a threat to online poker as people make it out to be imo, except on sites which are completely terrible at catching bots, and at anonymous tables. This bot plays a strategy too complex for people not to identify it as a bot, if your using like 6 different 3bet sizes preflop, it becomes quite obvious. So on sites like pokerstars, it won't be able to go undetected.

(again, not due to pokerstars bot catching team, which is fairly incompetent from my experience, but due to regs reporting it, providing evidence).

The real threatening bots, imo, are the ones that are 'real time assistance' bots, using pio as a search database.

This ai kind of has a base strategy similar to a pio search database, and uses realtime solving to come to more accurate solutions (and this allows it to add bet sizes not in the original strategy)

It doesnt need to use a ton of sizes everywhere. It could resolve the spots with whatever sizes they wanted.

This software is more similar to someone running pio/monker on the side while playing and if solvers could run solutions in real time.

I imported the HHs and looked at a bunch of spots, compared with pio and I’d be really surprised if it was more accurate, it seems way less accurate, there is a hand SBvsBB where pluribus peels A4ss pre, flop JT7r, flop goes check check, turn Ad which puts a diamond draw, sb bets 3/4ish pot, pluribus calls, river Ac, sb bets 8bb into about 10bb, pluribus raises to 60bb, sb jams and pluribus calls, it’s a fairly signicant mistake

it also oddly seems to have only 1 sizing between minbet and pot and its half pot, has a bunch of overbet sizings but no small bet or medium (75%) bet

Dealt to Bill [9c Ts]
Dealt to Pluribus [Qc Qd]
Dealt to MrOrange [3d Ah]
Dealt to MrPink [4s 7d]
Dealt to MrBlue [2h Qh]
Dealt to Joe [Qs 8h]
MrOrange: folds
MrPink: folds
MrBlue: folds
Joe: folds
Bill: calls 50
Pluribus: raises 200 to 300
Bill: calls 200
** FLOP ** [4d 5d 8s]
Bill: checks
Pluribus: checks
** TURN ** [4d 5d 8s] [3h]
Bill: checks
Pluribus: checks
** RIVER ** [4d 5d 8s] [3h] [Js]
Bill: bets 400
Pluribus: calls 400

???

there is always a lot of flop check backs in these limp/call spots, this line doesn’t seem too out of the norm

Looked through some hands..

what do you guys think of this hand:

Pluribus opens CO (I believe) JTcc, BB defends.
Flop - Qc3c2x
Pluribus bets 1/2 pot
Turn - 9x
Pluribets Pot
River - 6x
Pluribus checks behinds?

Seems weird..

Sure, meant the river flat. Std with 100bbs?


JT is kinda bad for continuing on river (blocking the kj kt club folds)

I think its close against villain’s sizing, I can get behind just a flat, its a hand we get pushed out of when villain bets/3bets, and not that many worse hands are calling a raise although a few should

also from what I’ve read it was choosing a node closest to the sizing used from villain, perhaps it was reacting to a bigger sizing than the one it was actually facing

Yes but still, I am pretty sure Pio bluffs the river a lot as we don‘t really have worse hands on the river..

PokerStars Hand #31048: Hold'em No Limit (50/100) - 2019/07/11 08:37:28 ET
Table 'Pluribus Session 31' 6-max (Play Money) Seat #6 is the button
Seat 1: MrWhite (10000 in chips)
Seat 2: Gogo (10000 in chips)
Seat 3: Budd (10000 in chips)
Seat 4: Eddie (10000 in chips)
Seat 5: Bill (10000 in chips)
Seat 6: Pluribus (10000 in chips)
MrWhite: posts small blind 50
Gogo: posts big blind 100
*** HOLE CARDS ***
Dealt to MrWhite [3s 2d]
Dealt to Gogo [7s Kd]
Dealt to Budd [Js 6c]
Dealt to Eddie [Kc 9h]
Dealt to Bill [9s 2c]
Dealt to Pluribus [Qs As]
Budd: folds
Eddie: folds
Bill: folds
Pluribus: raises 125 to 225
MrWhite: folds
Gogo: calls 125
*** FLOP *** [5s 7c 8d]
Gogo: checks
Pluribus: checks
*** TURN *** [5s 7c 8d] [Ad]
Gogo: checks
Pluribus: bets 250
Gogo: raises 850 to 1100
Pluribus: calls 850
*** RIVER *** [5s 7c 8d] [Ad] [Ts]
Gogo: bets 3400
Pluribus: folds
Uncalled bet (3400) returned to Gogo
Gogo collected 2700.0 from pot
*** SUMMARY ***
Total pot 2700 | Rake 0
Board [5s 7c 8d Ad Ts]



nitty for a bot . interesting it folds here considering it loves to call

Dealt to Eddie [Ac Jd]
Dealt to Pluribus [3s 3c]
Pluribus: raises 125 to 225
Eddie: raises 875 to 1100
Pluribus: calls 875
*** FLOP *** [6d 5d 6s]
Eddie: bets 690
Pluribus: calls 690
*** TURN *** [6d 5d 6s] [Kd]
Eddie: bets 1230
Pluribus: calls 1230
*** RIVER *** [6d 5d 6s] [Kd] [9s]
Eddie: bets 6980
Pluribus: calls 6980

Wow you're right! I applied some filters on PokerTracker and it there isn't a single hand where Pluribus bets at any sizing under pot that isn't 50% pot in a single raised pot. Obviously this is a huge flaw.

Take for example a hand like this:

PokerStars Hand #99033: Hold'em No Limit (50/100) - 2019/07/12 03:30:33 ET
Table 'Pluribus Session 99' 6-max (Play Money) Seat #6 is the button
Seat 1: Bill (10000 in chips)
Seat 2: MrBlue (10000 in chips)
Seat 3: Pluribus (10000 in chips)
Seat 4: MrPink (10000 in chips)
Seat 5: Eddie (10000 in chips)
Seat 6: MrOrange (10000 in chips)
Bill: posts small blind 50
MrBlue: posts big blind 100
*** HOLE CARDS ***
Dealt to Bill [Ts Jd]
Dealt to MrBlue [Qh Td]
Dealt to Pluribus [5c 5s]
Dealt to MrPink [Tc 7s]
Dealt to Eddie [Th 9s]
Dealt to MrOrange [8d 7c]
Pluribus: raises 187 to 287
MrPink: folds
Eddie: folds
MrOrange: folds
Bill: folds
MrBlue: calls 187
*** FLOP *** [6c Kc Ks]
MrBlue: checks
Pluribus: bets 312
MrBlue: folds
Uncalled bet (312) returned to Pluribus
Pluribus collected 624.0 from pot
*** SUMMARY ***
Total pot 624 | Rake 0
Board [6c Kc Ks]
PokerStars Hand #99034: Hold'em No Limit (50/100) - 2019/07/12 03:30:34 ET

PioSolver advocates betting this flop basically 100% of the time at a small size (25% pot in the sim that I ran) and it's easy to see why. MrBlue's range is quite polarised and has a huge amount of trash which he will fold to any size. Pluribus has a huge range advantage overall so Pluribus can protect it's equity and get value from pairs well with a small bet. However, MrBlue has a decent amount of kings and the top of his range is quite strong so betting large or even at 50% pot doesn't make much sense (EV difference isn't huge but definitely significant).

I think Pluribus is making quite a clear error if it bets all hands at 50% on this flop.

Preflop had to be trimmed considerably, I don't think there's much to read into preflop sizings.

yeah, the bot is not doing very good sizes, or not as good as what people have discovered in pio. The JTcc hand posted above would be a fairly normal overbet as well, but it just pots it. There is definitely some ev that it's losing out on with these limited sizing options.

There is no problem with the AQss hand, i don't get what's so bad about it? we have to analyze it's strategy, not just singular hands, there's nothing interesting we can say about that hand since we don't know what it's folding frequency is, but we do know it should randomize some folds there are at least sometimes.

Looks like a human would not play A4ss this way but how do you know the hand is a huge error? Im guessing you ran it in pio?

is this the entirety of play or just a single sample they released?

this seems like statistical tie controversey all over again

Yeah these were the only 10K hands played in the 5 humans, 1 AI challenge. There were another 10K played with 5 AI and 1 human but the hands aren't published for this.

Too small a sample to really say much but then the researchers claim they have some kind of variance reducing techniques which show the AI to be winning. Seems stupid if the sample is so small that the bot actually lost at 7bb/100 and but then this can come out at +4bb/100 with variance reduction techniques...