What will GTO look like when yer dead?
Or to put it another way: what will GTO look like in 50 years time (when (most?) of us are dead?)
Does it keep complexifying? Is there a finite point beyond which the boundaries of space and time prevent a type 0.7 civilisation from nashing?> eg. the practically available energy of our star?
Or.. will it be mostly the same, or sameish, but with better graphics ‘cause, meh, a % of the population will always be arsed to learn GTO up and to the point that they can print $$$$ from rich people (but not necessarily any further) - just like it is now?
*For the sake of argument I’m assuming money still exists, and poker is still a popular pastime. In fact: poker’s doing well because the live streamers are killing it in neuralink
10 Replies
there will probably be more precise solvers because people enjoy autism, but i think a lot of the new tools will have to do with stuff other than better gto, maybe more practical gto instead
Just a few things off the top of my head. I think we'll see most of these in the next decade:
- Agree with aner0 about a more practical focus. Tools that can show you the tradeoff between easier implementation and exploitability. AI coaches and built-in explanations of the strategy.
- The underlying algorithms will shift from manual calculations like CFR to neural networks (way faster).
- Nash Equilibrium Refinements. Better handling of ghostlines. Optimized solutions for ICM/Rake/Multiway spots.
- Connecting mass data to exploitative solvers. We'll probably see Bayesian game theory solvers that model population tendencies.
- Better abstraction techniques. You probably won't need to define the betting tree. It will just find optimal sizes for you.
- Better cheat detection.
- Better tournament equity models. The ability to instantaneously solve extremely large tournament fields (like 10k+ players).
- Alternative utility functions that can trade edge for lower variance.
- Postflop bunching.
- Tools that can solve arbitrary versions of poker with whatever rules the user implements.
I'd also bet that the training is way better. Right now it's hard to synthesize the information from a solver. I imagine in 50 years we'll have tools that optimize your training routine based on performance.
also in 2074 people will be crying about how poker is dead and it was so good in 2051, where they hadn't invented AI bunching exploitative algorithmic coaches yet
Damn. Kinda wanna stay alive now to play with the cool tech.
I think something could be done with the standard range grid. (or whatever the f it’s called)
It bothers me that the 2d version has inconsistent relative sizes of combos for offsuit/paired/suited, which are all geometrically equalised in the standard model. But if it went isometric (as in the video game sense) instead, i.e. AQo is ‘higher’ than AA and even higher again than AQs in block sizes, then it would be easier to navigate the disparities mentally?
Big problem being you’d have to invert offsuit with suited so that it was now on top/above, which would confuse everyone. But maybe you could flip it horizontally so that the 3d ranges were distinct.
2d ranges would eventually be known as ‘pancake ranges’ and generally thought of as lame, too abstract (you’re using pancake ranges bro – no wonder you’re overpunting diamond combos!) and much too imprecise to be classed as ‘proper GTO’.
I would do away with the 13x13 grid (which would make it applicable to any poker variant) and generate strategies that can easily be expressed in natural language
You'd probably optimize for
- Average EV loss vs equilibrium and a few simplified strats
- Practical exploitability, found by only allowing villain to make the type of broad strategic adjustments humans can
- Simplicity
So not the current way where villain knows your exact strategy and can make absurdly complex adjustments
some kind of neuralink technology where the solver is embedded into your brain. True real time assistance, that will make "studying" obsolete.
"Humans are the sex organs for the machine world"
I think something could be done with the standard range grid. (or whatever the f itÂ’s called)
It bothers me that the 2d version has inconsistent relative sizes of combos for offsuit/paired/suited, which are all geometrically equalised in the standard model. But if it went isometric (as in the video game sense) instead, i.e. AQo is ‘higher’ than AA and even higher again than AQs in block sizes, then it would be easier to navigate the disparities mentally?
Big problem being youÂ’d have to
I agree the standard range grid makes it difficult to visualize things. There are so many more combos of offsuit hands, but they take up the same amount of space on the grid as the suited combos. I'm guessing the solvers of the future will have a lot more graphical interfaces that you can choose between, maybe some 2D and some 3D.
I would like to see some sort of graphical interface that maps all the hands in terms of equity, with higher equity hands on top and larger area sizes indicating more combos. Color coding could indicate preferred actions with each hand. The last part would be similar to current models.
Then you could overlay your opponent's range over top, with it being semi-transparent so you can see both ranges.
This would make it easier to understand GTO strategies. You could see when your opponent has a large area of hands way down on the chart compared to your hands, which makes a small range bet attractive.
You could see the relative surface area size of your value hands, along with a corresponding surface area of your bluffs. This would make it easy to see spots where it would be easy to overbluff, for example. Like, "Oh the value portion of my range is really small here and I have a ton of potential bluffs. I need to be careful not to overdo it."
Another possibility would be a chart that has general equity and nut potential on the two axis. The high equity low nut potential combos would perform better when ranges stay wide to the river, so a more passive line might be preferable. The high equity high nut potential hands would want to inflate the pot. Low equity low nut potential hands would be trash that you would generally be folding and eliminating from your range on future streets. Low equity high nut potential hands might make good semi-bluffs. You could color code the preferred plays with all of the hands plotted on the chart so you can visualize general trends.
This type of chart would make it a lot easier to understand common properties of seemingly different types of hands that want to play similarly.
Also, AI will likely get much more advanced and better at breaking down anything you want to know. Maybe it will be able to create charts like the ones described above in real time just based on a simple verbal description, with the AI asking follow up questions as necessary?
An analogous to nodelocking feature but general for future tendencies. E.g. we can't tell a solver that villains overfold all rivers 5% (or overfold a particular river type x% of the time, overcall other y% and play optimal the rest etc), so the solver can give preflop, flop and turn optimal counterstrategies considering villain will make z type of mistakes by the river, etc. Same for turn mistakes, giving accurate pre and flop deviations.
Linking all of this to mass database softwares, and the solver being AI-based.
I think this will happen while I'm still alive, maybe the next 5 years at most.