Pluribus Poker
Posted : admin On 7/31/2022'Pluribus achieved superhuman performance at multi-player poker, which is a recognized milestone in artificial intelligence and in game theory that has been open for decades,' said Tuomas Sandholm. Pluribus beats top poker pros at 6 max The poker world continues to be spooked by the advancement in artificial intelligence and the impact it could have on online poker.
Ai Texas Hold Em
GitHub - whatsdis/pluribus: Implementation of Pluribus by Noam Brown & Tuomas Sandholm, a Superhuman AI for 6-MAX No-Limit Holdem Poker Bot. Researchers say they have designed a bot called Pluribus capable of taking on poker professionals in the most popular form of poker and winning. Four-time World Poker Tour title holder Darren Elias.
Recently, some truly groundbreaking news hit the poker wire when it was revealed a poker bot had been developed that could crush some of the top human players in six-max no-limit hold'em. Pluribus, as the AI program has been dubbed, defeated lineups including Darren Elias, Tony Gregg and Nick Petrangelo among other notable poker pros.
The news represented a significant breakthrough in artificial intelligence development, and could have huge implications on poker at multiple levels. For example, online heads-up no-limit hold'em has become essentially a dead game in the aftermath of an earlier bot, developed by the same team, demonstrating the format was machine-solved.
Does a similar fate await online six-max? Will online poker sites need to up their bot detection techniques by a few orders of magnitude in light of the revelation that Pluribus runs on about $150 worth of CPU power?
While these are valid considerations, this article will instead delve into some of the strategy demonstrated by the unusual lines taken by Pluribus in some of the hands posted to YouTube and Facebook by the AI team. All hands are played $10,000 deep at $50/$100, and all thoughts come with the caveat that I haven't been a regular in online six-max cash since Black Friday, 2011.
Hand #1: Pluribus Shoves Many Times the Pot
Hand Details
In this hand (shown above), Player 3 raised under the gun to $225 with . Pluribus called in the next seat with . Player 1 then called from the small blind with and Player 2 in the big blind called as well with .
The flop came and both blinds checked. Player 3 put in a bet of $300 and Pluribus shoved all in for $9,775. Everyone folded and Pluribus dragged the $1,200 pot.
Thoughts and Possible Implications
This hand is notable not because of the action Pluribus took — raising a small continuation bet while deep-stacked after flopping a combo draw — but because of the massive sizing Pluribus used.
Starting preflop, things get going in a bit of unusual fashion right off the bat. Flat-calling against an under-the-gun open isn't really a standard play in a tough six-max lineup. I'd consider it a better hand to raise with as a bluff, although that works a little better when you're deeper-stacked. Flatting here opens the door for aggressive players to squeeze behind you, as well as to come along with cards that could dominate and stack a six-high flush or straight.
Apparently, small suited connectors can indeed be played for a profit with a call in this spot if you're good enough postflop.
Pluribus Poker Bot For Sale
After the flop, Pluribus finds a fairly strong combo draw in a multi-way pot. When the preflop opener makes a small continuation-bet of $300 into $900, Pluribus has options. It can call and play in position or it can raise as a semi-bluff.
I prefer raising as this hand has zero showdown value without improvement, and a raise is more likely to get the pot heads-up where the hero can win with a bluff or even something like a pair in some spots. I'd probably make it something like $1,400 in this spot, hoping to fold out the blinds and set up a possible triple barrel.
What Pluribus actually did — shove all in for more than eight times the pot — shocked me. How can this be optimal, I wondered?
While my simple brain, capable of only beating small-stakes cash and tournaments, can only offer conjecture, it's interesting to think about possible reasonings for this play.
First, the pot is multi-way, which represents the only scenario where can be in real trouble — e.g., someone has a strong pair or a set and another player has a dominating draw like . Shoving gives Pluribus the maximum chance to get heads-up or win the pot right away.
Second, perhaps Pluribus, through its millions of trials playing hands against itself, has determined that the optimal play against strong opponents is for Hero not to expose himself or herself — or itself — to tough decisions on later streets with a strong draw.
Consider what would happen, if say, Hero made it $1,400 as I posited. Many difficult scenarios can now develop.
Maybe someone in the blinds puts in a cold three-bet, maybe to something like $4,000 or all in. Either way, Pluribus has six-high and no fold equity with the possibility of being a decent underdog to a set or a strong draw.
Maybe one of the blinds cold calls and the opener calls. Hitting the draw may or may not win the pot now, and Pluribus can't confidently get stacks in.
Even if it just folds back to the opener, and he calls, unless the draw comes in on the turn, it's not going to be clear what the best play is. Pluribus could check back, but then getting paid on the river may be difficult. If Pluribus fires another barrel and is called, it sets up a river bluffing decision on another brick.
It seems clear that it can't be a mistake to play the hand this way. The only question is how the hell does one balance in this spot shoving eight times the pot? My best guess is that Pluribus has figured it simply doesn't matter. The hero picks up a $1,200 pot enough of the time that it makes up for any equity lost when called by hands that are slight favorites.
Next week, we'll continue to examine other unique lines from Pluribus and see if we can glean anything of import.
Tags
cash game strategyno-limit hold’emmulti-way strategydrawing handsfold equityequityaggressionAIpoker botsPluribus
So after looking at Pluribus poker hand ranges at all positions and Pluribus bet sizes, what learnings do we really get ?
Pluribus Poker Github
So let’s look at the rfi frequencies actually played by the 10 pros and Pluribus on the course of 10k hands. Bear in mind these are the frequencies actually played, so slightly different from the actual ranges (although very close) ie when Pluribus gets dealt a lot more JT than 32, it does have an impact on the frequency. So not the same numbers that I’ve shown in the ranges.
Nothing out of the ordinary here, except the small variations between players show there is not yet a full concensus on rfi preflop ranges, but I think this is coming soon. Basically the Pluribus poker hands opened ranges are very close to snowie or upswing poker ranges, so this will converge.
Key Learnings RFI Strategy from Pluribus Poker Hands
Pluribus Poker Ai
- Ranges don’t change much, review, fine tune ever so slightly, add a few Kxs early.
- Randomise bet sizes, check the charts for actual sizes per position, nothing too complicated. 50% 2BB, the rest split between 2.25 and 2.5. Exception from BU with more 2.25 and 2. 5 than 2BB.
- From the SB, 63%range, then 50/50 limp/bet, and same split between bet sizes.
- Once in a blue moon for fun, bet 3, 3.5 or even 4BB. Any handx doesn’t matter.
- Not saying this is what we should do, just the poker hands Pluribus plays and their bet sizes, but if past history is anything to go by, decent players will move that way.