Quote:
Originally Posted by spaceman Bryce
Can anyone tell me what gto wizard actually is? Ive only looked at it briefly and could be wrong. It seems like a browser based piocalculator lite for small stack sizes. What language was it programmed in? does it work with sql ? What other games can you use it for? What kind of combinatorial things can it do? does it have a scripting language? Why would someone use it? It appears, and I dont gave a premium membership , but it appears limited. As far as I know you wont beat high stakes with it, or any stake optimally.
The core product of GTO Wizard is a web-based wrapper around Piosolver. As in, they ran a lot of Pio solutions, and put them in a nice webapp for people who don't want to have windows, buy piosolver, run all the spots themselves, and in some cases perhaps lack a powerful enough computer to do so.
There's basically two high level classes of solvers: reinforcement algorithms and neural nets. Reinforcement learning tools are ones like Piosolver and Monkersolver - to make a chess analogy, they are basically algorithms that "look ahead" moves to see how it works out. Neutral net tools, on the other hand, take in a bunch of "features" of the current state of the game, and do a big fancy linear regression using a lot of existing examples to train on. As you can imagine, the "lookahead" solvers are more technically accurate but the neural nets can be much faster.
The supposed best poker AI in the world Pluribus built by Facebook uses a combo of those approaches.
PokerSnowie was the best known neural net solver but their marketing presence seems to have died a bit. Ruse AI was a team of PhDs in Canada doing their own neural net solver and GTO Wizard acquired them so they can now solve arbitrary spots using Ruse.
Since it's a webapp, the frontend is written in Javascript (or Typescript) , and based on their job ads, the backend is written in Python (with the Django framework).
GTO Wizard does not expose any sort of SQL interface which would be very unusual for a web app to do so directly. What would be usual would be some sort of HTTP API, which they explicitly and proudly note they don't offer because they don't want it to be used for RTA.
It currently is just hold'em.
As far as why someone would use it, it's the same use case as Piosolver which is basically if you study the game tree at equilibrium you will develop a better underlying understanding of the game mechanics. Learning how often the solver bluffs, which hands it picks, how often it calls, can be illuminating even if you don't copy it directly. It would be extremely difficult to copy directly since it uses more bet sizes than a human could implement and your opponents are certainly not going to stick to a few fixed bet sizes. But it's pretty clear that solvers are a powerful educational tool as the vast majority of the best players attest to using them as a study tool, even if that's still a controversial topic on reddit and 2p2.
I'd argue that while you're unlikely to beat high stakes by just studying equilbrium, you're very unlikely to beat high stakes without studying it unless you are a true savant.
Both piosolver and gtowizard also allow "node locking" which lets you take a node in the game tree, and instead of doing the equilibrium strategy, you force it to pay differently (most often, you force it to play how you expect your opponent actually plays) and you can see what the optimal exploitative strategy is against that adjustment. However, node locking is clunky because the game tree has so many nodes. Often, if you lock an early node such as a flop decision, it grossly overcompensates on later streets in ways that won't be natural and your opponent won't do. As a result, Piosolver introduced "incentives" which _nudges_ decisions a certain direction rather than "locks" them which the Piosolver team produces better results.
Your general line of questioning seems to indicate you have a more old school attitude to studying with tools like OddsOracle. That stuff can still be very useful as a tool used in conjuction with solver study, though I do think solver study more directly answers a lot of questions that you're indirectly getting at with the old-school approach.
I think there is still a _lot_ of room for improvement i the poker studying tool market. You can learn a lot from looking at pio or gto wizard but there's obviously still a huge bridge from a solver solution to understanding it and applying it. This is why I'm building my own product in the space , a GTO-Wizard competitor of sorts. However, I have to acknowledge that they put together a fantastic product, particularly in terms of UX and UI design, as well as having uploaded a ton of solutions for various spots, so are a very scary competitor, and I'm mostly hoping I can carve out a corner of the market rather than "win" it.
With that said, IMO if you're a hardcore poker nerd you're still better off with Piosolver over GTO Wizard as it's a bit cheaper amoritzed over 1 year+, and can solve arbitrary spots with a reinforcement learning solver which is more accurate, and do things like run your own aggregated reports on custom spots, and be on more bleeding edge features like incentives. The Piosolver team is also much more open to being scriptable and has documentation on how to run Piosolver programatically, opening up a top of options. Their Discord is also less chaotic and it's easier to chat with their team. However, most people don't want to buy expensive big Windows machines and run all these custom spots, hence why GTO Wizard has taken off, along with a massive marketing push by them.
