Hey, just some thoughts on how to beat a GTO bot.

Say we can find a perfect amount of BB to play vs a bot. Say its 44, 67, 99, 131 or some arbitrary number where the math is skewed to some decimal point in your favor if you are going to fold or shove a range of hands. You mathematically determine some hands that you are going to fold, and then some you'll shove. There has to be some BB number where you can win some small amount (%) against the bot (there would be some BB where you would lose to). So we just have to find the ones where we will win.

This is assuming that GTO (math) is inferior to math plus strategy which you can find a long post of mine here.

https://forumserver.twoplustwo.com/1...rfect-1797207/

Thoughts?

so, gto bots don't and can't exist, because gto is not known and can't currently be calculated.
There are pseudo gto bots that calculate one abstraction or model of the game, but it will be missing certain stack depth and sizings calculations.
This is not due to some limitation of math versus math+strategy or whatever oogabooga that is. It's due to computational limitations that basically don't allow us to solve the entire game from A to Z.

The only way to beat a pseudo-gto bot would be to use sizings or play at stack depths that haven't been calculated by their model of the game. This will depend on how the specific model of the specific bot was calculated. As long as you use sizings that are part of the model, you can't really win unless it's a really poorly made bot.

Deep cfr can handle variable stack sizes

Search can handle variable stack sizes and bet sizes

You won't be able to exploit a strong ai by using a weird stack or bet size

neural network solvers aren't nearly close to gto yet, they output some weird ****. i have only messed with deep solver so far though, maybe some other one is better, but probably still not "real" gto.
Of course when i say you can exploit a gto bot due to its innacuracy, i'm really just playing devils advocate. In practice, it's impossible for a human to beat a gto bot, even a not so great one.

ya aner0 i basically know gto cant be solved and if you want to know about my oogabooga check my other post. But if a bot played the math perfectly, you might be able to win 50.00001% of the time using the shove or fold technique analysed for every hand I think. Basically im using GTO as the "math" element of the game, and strategy as the "unknown variable" element of the game. In my long post I tagged i basically went into why poker should be considered correctly 50% math and 50% strategy for humans to correctly consider when making a move, I don't know if its because we have a left brain and a right brain but whatever. Just practicaly speaking as opposed to some other weird way of going about it.

But ya I understand GTO is not solvable unlike chess, which is all math, or opposed to rock-paper-scizors which is all strategy. Basically everyone has a happy medium of whatever combo the use, in games with a gray area, sorta like gray matter.

But I think this strategy could work if someone did solve the GTO aspect of all combined tethers and analogues of every situation, non factoring unknowns. idk

Can you explain the difference between "math" and "strategy" as you interpret them?

In your original post, I argued that all strategy can be reduced to math, and you got quite angry with me. So would you mind explaining in concrete terms why you think otherwise?

Rock-paper-scissors can be completely reduced to a very simple math problem. Chess is not solved because it's too complex, engines just make good approximations. But again, it is reduced to a math problem.

-----------

A shove/fold strategy would definitely lose a ton money against any GTO bot with sufficient complexity. The only exception would be when stack depth is very low and the optimal strategy becomes push/fold anyway, in which case you would tie against the bot with perfect play.

lol not anger dw. strategy involves the entire spectrum of a particular stimuli. math is a static integer essentially in that set.

I think in poker its essential to diverge the concept of thought into these spectrums because you have ranges and hands. If we were talking about rock paper scizors i think for practicality reasons (not objective reasons), you could converge into one idea spectrum.

Since bots will never be perfect, its important to keep the idea of strategy alive so that one can out maneuver the inherent analyzed game trees in a continuous fashion relating to hand vs range aspects. anyway

I still don't understand. What do you mean by the "entire spectrum of a particular stimuli"? What is it stimulating? Can you give me an example?

Math can be represented with variables, not just static numbers.

basically were all poker players here. you play 1 hand differently if you have 1 hand to play then if you have 10k hands with the same player. Me personally I would tag this down to strategy since its used that way in english, but if you were to say he did the math in his head and came out with that particular answer then w/e.

Trying to reduce things that dont need to be reduced to a single variable is counterproductive I would argue if it not particularly over-conceptualized to start with. You might argue that a computer only thinks in terms of math, but it thinks in binary. You need lots of shading in computers to work proper code.

your other thread is brain rot, just as this one.

gto in poker is solvable just as chess is, it's just it takes a looooong time to solve with current computation (not possible in practice yet). There's no distinction between math and strategy.
Solvers don't do math equations until they come up with the best strategy, they actually simulate both players designing strategies that counter each other until the uncounterable strategy is arrived at.

just stop talking out of your ass i have covid and it makes my head hurt even more

that doesnt even make sense kid, if one counters each other then its going to be a +ev play infinite is a long time. actually its infinite. my bad. continue with your covid.

I think my point has been made, dont look to close. you might touch your nose to the computer.

it's not, because the solver simulations are not just trying to exploit each other, but also to keep their ranges balanced by mixing combos as to not get exploited. This arrives at equilibrium which LITERALLY means neither player can deviate and make money. By definition.

i fail to see your point. bigger picture stuff

point is perfect gto can't be exploited because your best exploit is just indifferent (same EV) to playing gto yourself. if it can be exploited then it's not perfect gto

not 50.00001% or 49.99999%. Exactly indifferent.

Checkers is solved. i agree a game can be solved. I also agree with the point that a game with unknown variables will always have a strategy component to it. The way we say GTO in poker is more like game theory probable. If you take the amount of what percent a game is math vs strategy it will be 50% both if you are balanced. maybe not in practicality but in terms of what we can talk about its definitely equally viable options to consider. Think poker. this isnt a computer programming class.

If bots are 99.999x100 right. You could use this same math based calculation for a strategy to find a certain amount of BBs that a shove fold strategy would be winning against (i think, this is why i wanted thoughts on it) and other it would lose against. say 0.00001%. In the whole strategy it would be equal because it gto but that doesnt mean you cant lose a hand.

i don't think half of your sentences are even coherent

well i was gonna wish you well with covid, but hell.

maybe i'm just dumb as a rock but i legit can't understand half of the things you're saying because you're using these salads of random words. just speak in plain english for dummies plz

let me try once more. my original thesis for this thread is that poker bots will have a weakness for a particular range of hands for a particular set of bbs. sound fair?

in theory you could never know which bots have which calculations covered using like terabytes of info but if you did know it, I think one bb would be acceptable, even if it was like 10 billion four hundred thousand and four big blinds. its like the infinite hotel problem covered by veritasium on youtube. it just converges at a smaller point.

There's a lot of wishy-washy nonsense going on here.

I asked you to explain the difference between strategy and math:

A GTO strategy plays the same way regardless of whether you play one or one thousand hands against some player. You can talk about dynamic strategies that are designed to exploit mistakes and adjust on the fly; but then you're no longer talking about GTO.

Any strategy can be represented mathematically, regardless of whether that strategy is dynamic or static.
-----------

As far as GTO goes, you can't exploit true GTO because then it wouldn't be GTO in the first place, by definition. You can maybe exploit inaccuracies or limitations of the solution, with absolute GOD level play and an infinitely complex strategy. Similarly, you can beat a chess engine with GOD level play. But this is well outside the mortal realm.

A braindead push/fold strategy will lose horrifically in the long term.

------

Let's design a HU strategy. You tell me the effective stack depth, and I'll solve the call/fold against that push. We lock that call/fold in, and then you can try to find a push/fold range that increases your EV.

Yes, at the time being where a solver can't work in real time to solve new spots, you could find holes in the model of the bot. You would have to play very very well on that uncovered spot though to exploit the solver hole.

In the future, solvers will be fast enough to work in real time and if you use a different sizing or stack depth it will just solve it on the fly.

Even this is getting way into the weeds because in practice, the solver will play those "holes" in their strategy probably better than any human could still, even if not perfect.

hmm so youre saying GTO combines hand and range. I think it is playing hands within a range, which is a little different, there is still a distinction imo.

As to your last question, Id need to know the practicalities of the computer code to figure it out

Yes, GTO bots work with a range of hands, they don't just play a single hand in a vacuum.

Here are the parameters:

• We start 100bb effective.
• You post 0.5bb, I post 1bb.
• You act first. You can either shove or fold.
• There is no rake.

Here's the GTO response against a shove. You can see it calls with a rather tight range. Let's imagine it will always call/fold using this strategy, regardless of how you play.



The bot calculates that with perfect play, you will lose 43.35 bb/100. That's because a push/fold strategy sacrifices your positional advantage postflop.

Now, I would like you to try and calculate the optimal push/fold strategy. Can you win more than - 43.35 bb/100? If so, you'll beat GTO.