Some computer scientist at university of alberta have made a GTO bot for 200BB poker heads up NoLimit using huge computing resources. Does it mean an average Joe can solve 25BB heads up nolimit poker on his laptop ? As this has much fewer game states. What about 37.5BB?
It's a very complexe question, The short answer is no, And I doubt that there are many people on this forum that can answer you that. I mean even simple/pio solvers are far from being "perfect". You see bots are exploitabale, what ever other people say, they luck a lot of things. They are playing with approximations, farther from the flop you move, the more they struggle. They make a lot of errors on the river, so when you find them, you can take advantage of them. They tend to overcall/overbluff/undercall/underbluff a lot. And you may think that the owner of the bot may repair them when he spots them, but that's not true, since if he repairs one error, another will appear in some different spot. Let's say someone cbets too widely on the turn, you make it closer to perfection, but then, he might have a weak checking range, and this goes on and on and on. However, with a human behind them monitor that will use the help of the bot without leaving him to play solo, the bot is unstopable. So no, an average bot can't beat no limit holdem even at 25bb and less. What you were citing in the first place is libratus, and it's a huge bot, and a bot like him will never appear in poker since you can make much more profit using it elsewhere, let's say in trading, it's just not worth it.
I didn't think that bot was playing GTO let alone achieving GTO. I thought it studied the specific 4 opponents after each of the 20 day sessions and worked on improving strategies vs what their opponents did. I could be wrong though, but that was my impression from reading the way it was setup. It did extremely well in the 2017 setup, though I'd imagine both human fatigue and computer adjustment played a large part the longer that went on.
In any case, that is an amazing supercomputer built for one specific purpose. It was also the result of years of study and prior (failed) attempts.
I agree with Meroving, the short answer is pretty clearly no.
No this bot Deepstack was playing GTO. Its not exploitable by any human in a lifetime for more than a dollar. The paper they published included a proof of this. If you ant to make a bot all you have to do is read their paper and implement it, its all described there. Now how many BBs you can solve for depends on your computing resources, 25BB is doable on a regular dekstop. For spin and gos you need more resources.
