Learnings from Pluribus poker AI 10k hands

Carnage Mellon University published in Science magazine the 10 000 poker hands played by Pluribus in 6 max no limit holdem against 10 pros.

Indeed the impact is so important that the researchers have signed a major contract with the military to use the findings.

Poker AI military contracts

Anyway, what I care about is the poker. So I input all the poker hands in my database to gather information on how Pluribus plays: ranges, bet sizings, defense frequencies, flop play in position and out of position. I found a lot of info and decided to publish it here. I hope it will be useful to others.

And considering previous AI advances, in particular Deepstack and Libratus, have revolutionised poker strategy with game theory, I’m certain players will change their game to take into account Pluribus poker results.

Players will start varying their preflop bet sizes. Small blind open mixed strategies will become much more standard (mixing limp and bet) etc.

Of course I could rather be wrong, but I think that whatever Pluribus does that is not too hard to do, many many human poker players will start incorporating into their play. I’m trying to find all these here.

I’ll be updating this site every week. For now you can find:

Openings: poker hands ranges and bet sizing