How far has NL poker come? How far do we have to go? (abstract/long)

[VanVeen]

hrm, i'd say the biggest leaks are related to hand range strength evaluations given preflop ranges, the community cards, and the preceding actions in the hand. not bluffs or snapping off bluffs specifically.

ideally you'd assign an integer to your hand expressing relative strength that's a function of your opponent's preflop range, the community cards (which would incorporate the likelihood of future community cards), and the actions preceding the current decision node. if it exceeds some value you act aggressively; if it exceeds some value but falls beneath another you call, etc.

i don't think many players: a) assign relative value correctly; b) determine the minimum and maximum values needed to call/raise vs. a specific opponent; c) adjust properly over time to variable strategies, i.e. they will change the minimum and maximum values needed to bet/raise/call far more than is rational.

in the not too distant future someone is finally going to do the obvious and create a pokerstove-like program that evaluates preflop ranges with different community cards. you'd plug in a preflop range, select community cards, and see a breakdown of each different hand type given the board. once something like that shows up i think the games may start changing a lot again.

[VanVeen]

kotkis,

this is unlikely to be true. unexploitable strategies can still be dominating, and it's a pretty safe bet that humans would unwittingly choose strategies that are dominated.

[HP]

[ QUOTE ]
[ QUOTE ]
Here's a question for everyone:

Suppose some guy figured out how to play HU NL in an un-exploitable fashion (or at least like, very, very close to un-exploitable)

Now suppose he sits at FTP 300/600 for 8 hours a day playing anyone who sits with him

what's his daily rate? how many of you are making more than this currently?

[/ QUOTE ]
It would be very close to 0ptbb/100 if not exactly that. Figuring out the NE strategy (ie. perfect unexploitability) is not about turning a profit, it's about breaking even.

[/ QUOTE ]

not necessarily true at all

I could come up with very simple counterexamples

[kotkis]

VanVeen,

Yeah I was about to include the chance of humans using dominated strategies to increase the NE strategys win-rate, but I didn't think this effect was significant enough when we're dealing with the types of aba and Antonius here. How much do you think aba could increase his win-rate if he was able to completely eliminate any dominated strategies from his overall game?

HP,

I didn't provide any examples, but I'm interested in hearing about yours.

[Triumph36]

[ QUOTE ]

in the not too distant future someone is finally going to do the obvious and create a pokerstove-like program that evaluates preflop ranges with different community cards. you'd plug in a preflop range, select community cards, and see a breakdown of each different hand type given the board. once something like that shows up i think the games may start changing a lot again.

[/ QUOTE ]

i wonder if you could expand on this - right now, to me, it seems a little too complicated to adequately inform anyone. do you mean to say that within a given range (say, x% of hands being raised on the button), it would evaluate your hand against that x% range? i'm still not sure this information could be used optimally - nor that it's particularly better than what's being done now.

To me it sounds like you would say below or above a certain opponent's range holding a given hand, you should raise, call, or fold - but you should be doing the former two less than 100% of the time anyway in a given situation, thereby increasing your possible range (and making your opponent therefore play worse).

[Micturition Man]

[ QUOTE ]
hrm, i'd say the biggest leaks are related to hand range strength evaluations given preflop ranges, the community cards, and the preceding actions in the hand. not bluffs or snapping off bluffs specifically.

ideally you'd assign an integer to your hand expressing relative strength that's a function of your opponent's preflop range, the community cards (which would incorporate the likelihood of future community cards), and the actions preceding the current decision node. if it exceeds some value you act aggressively; if it exceeds some value but falls beneath another you call, etc.

i don't think many players: a) assign relative value correctly; b) determine the minimum and maximum values needed to call/raise vs. a specific opponent; c) adjust properly over time to variable strategies, i.e. they will change the minimum and maximum values needed to bet/raise/call far more than is rational.

in the not too distant future someone is finally going to do the obvious and create a pokerstove-like program that evaluates preflop ranges with different community cards. you'd plug in a preflop range, select community cards, and see a breakdown of each different hand type given the board. once something like that shows up i think the games may start changing a lot again.

[/ QUOTE ]


I think the most frequent mistakes would be related to value betting/raising (or not value betting/raising), which I think is what you are saying here.

However I think the strongest current players, though they are no doubt far from optimal in this regard, would not actually be losing much value to an optimal bot here.

The reason I say that is it's only the marginal value situations that are hard to intuit, and these by definition have a low EV attached to them.

OTOH you may be right... those marginal value situations are extremely common, so that could add up to a big chunk of EV.

The reason I mentioned bluffing and bluff-catching is that I am certain that even the best players have certain spots where they are hugely exploitable, because they assume "Nobody ever bluffs here"... which may well be true. But an optimal bot *would* bluff there.

(The significance of human bluffing being off is just that people's opinions of how frequently other people bluff in certain spots is largely conditioned by how frequently they bluff themselves in those spots.)

Also to whoever said the bot would have approximately 0 EV against the best human players... this is nuts. I don't think the #1 and the #2 players, whoever they are, are really literally close to 0 EV against each other. How can the #1 opponent be even closer to a perfect bot?

(Especially when you factor in the fact that the unfamiliarity of the bot's play would wind up inducing bad adaptations (i.e. tilt)).

[HP]

[ QUOTE ]
HP,

I didn't provide any examples, but I'm interested in hearing about yours.

[/ QUOTE ]

call the un-exploitable player A, and the other player B

say there is X bb's in the pot, and 4X bb's left in the effective stack. Flop is:

A98r

B checks, A bets pot (or whatever happens to be the perfect size), B pushes with 43s

B should have at least had a pair or gutshot, giving him more equity when called. This has to increase player A's winrate

[kotkis]

Yeah alright, so you're talking about dominated strategies. Maybe I'm overestimating the level of play by Antonius, aba and co. but I would still estimate that the biggest mortal human player would still achieve the highest win-rate in the whole playerpool using exploititive strategies against his exploititive opponents compared to what the bot would be making against the overall playerpool.

[RepulseMonkey]

[ QUOTE ]
hrm, i'd say the biggest leaks are related to hand range strength evaluations given preflop ranges, the community cards, and the preceding actions in the hand. not bluffs or snapping off bluffs specifically.

ideally you'd assign an integer to your hand expressing relative strength that's a function of your opponent's preflop range, the community cards (which would incorporate the likelihood of future community cards), and the actions preceding the current decision node. if it exceeds some value you act aggressively; if it exceeds some value but falls beneath another you call, etc.

i don't think many players: a) assign relative value correctly; b) determine the minimum and maximum values needed to call/raise vs. a specific opponent; c) adjust properly over time to variable strategies, i.e. they will change the minimum and maximum values needed to bet/raise/call far more than is rational.

in the not too distant future someone is finally going to do the obvious and create a pokerstove-like program that evaluates preflop ranges with different community cards. you'd plug in a preflop range, select community cards, and see a breakdown of each different hand type given the board. once something like that shows up i think the games may start changing a lot again.

[/ QUOTE ]

A G-Bucks Stove you mean then eh ?

[w_alloy]

[ QUOTE ]
everyone playing casts about for an overall strategy that works. some can cast farther than others. some cast farther and faster and can test more strategies. the entire group of strategy seekers, however, all bumble around the same area of strategy space because they're all constrained by the same computational deficiencies. stupid people, small circle. smart people, somewhat larger circle enveloping the smaller. smartest, larger still. area relative to total area? still totally insignificant. there is thus a ceiling to how much progress can be made.

[/ QUOTE ]

This is completely irrelevant. To suggest that modern strategies are limited by their complexity is ludicrous. Players don't do all the required math and game theory at the table, nor should they. Players are capable of implementing much more intricate strategies than are currently in use, and these strats can be come up with using limitless resources away from the table. Did you watch Boosted and Jman 8 table heads up? How would this speed of play be possible if people were stretching their capacities on each decision?

[ QUOTE ]
no one is capable of even sampling the overwhelming majority of 'strategy space', even by chance.

[/ QUOTE ]

In practical terms, I think this statement ranges somewhere from very misleading to untrue. We have limitless time, and human abilities are a pretty insignificant limiter.

[ QUOTE ]

no, it will not converge towards game-theoretic optimal play because the selection pressure acting on the population do not favor those who deviate the least from theoretical optima. the onus is on you to demonstrate otherwise. strategies are selected for their efficacy at exploiting imperfect, irrational, and heuristically-inclined humans with decision-making biases out the ass.

[/ QUOTE ]

All these factors you list are being decreased over time, and are mostly close to zero sum anyways (for example, some people are too tight, some too loose, and these cancel eachother out to a degree). Even so, flawed strategies exploiting flawed strategies are more likely to lead toward unexploitable strategies than away. I don’t see how you can argue against this statement.

[ QUOTE ]

the devil is in the details. most of the real 'game-theoretic' progress (avoiding errors that are salient enough to be reliably exploited) that's going to be made is already done.


[/ QUOTE ]

The 1st sentence doesn't mean anything, and the second goes more to my point then yours. If some of the "real 'game-theoretic' progress" is still left to be made, then won't we move towards unexploitable ground as progress is made?

[ QUOTE ]

this post is a lot like my first one. i don't think you read it!

[/ QUOTE ]

I read it carefully. I think you are misunderstanding some key parts about the nature of unexploitability and evolutionary processes.

Play has improved towards unexploitable consistently over time since the begininning of NL. I'm sure there have been many hitches and backwards movements, but on whole play is worlds less exploitable than it used to be. The proof is on you to show that this overwhelming trend is cooincidence or fad, and that it won't continue.