Quote:
Originally Posted by RaiseAnnounced
Terms like "solving", "GTO", and especially "unexploitable" get very slippery when applied to multiway situations. Because EV is no longer a closed system between two players where a loss for one player necessarily means a gain for the other, some combination of your two opponents' strategy can make any strategy -EV.
In the most extreme scenario, two great players who are literally colluding with each other will make money against an equally great player playing honest. A neural network might come up with the best solution for what to do in the BU facing an open from a 18% HJ opening range from the best player in the world and a BB left to act who's the second best player in the world, but that strategy would lose money if the HJ is folding the bottom of that opening range but then opening ATC when the BU holds a premium. This isn't just a pedantic distinction, because any number of real-world scenarios against two opponents who aren't (purposefully) colluding are possible that would approximate this sort of situation.
I don't know much about programming, but AFAICT a program that solves spots by approaching an equalibrium where three opponents are unable to get an edge on each other is impossible. I also don't know much about neural networks, but it seems more possible to me that they would be able to come up with strategies that are optimal in either most situations or against the toughest opponents, while acknowledging these are still practical and contingent rather optimalities, not theoretical ones.
Nice post. Also those optimal multiway "solutions" where the sim assumes the other players are the toughest opponents totally break down when one of the players is playing sub optimally; not as the simulation expects them to play. In other words it wouldn't be a practical "solution" in a 3 way pot with one super tough reg and 1 super bad whale, even in the way that you approach your strategy vs the tough reg.