In Response to AI takes on poker players : No Limit H/E = Human Edge I think and sincerely hope, as it could kill the online game, that NLHE will remain hard for computers to beat. Posted by StayOrGo
I'v been watching this live and the bots are winning at 13bb/100 after 6k hands which is very scary!.... BUT
And it's a big but....
Every hand is played with 200bb stack regardless of previous hands so it is only trying to solve at one stack size.... if this was played to go super deep the bot would more than likely get destroyed
I'v been watching this live and the bots are winning at 13bb/100 after 6k hands which is very scary!.... BUT And it's a big but.... Every hand is played with 200bb stack regardless of previous hands so it is only trying to solve at one stack size.... if this was played to go super deep the bot would more than likely get destroyed Posted by Itsover4u
the bot isnt trying to solve poker, it is trying to beat 4 human opponents. this isnt a scientific venture into game theory; it is one upsmanship and a pseudo scientific ego war.
the previous bot, cepheus, was the same. it was reprogrammed between sessions to plug leaks and exploit the human team's exploitative plays. it even deployed different versions of the bot against different players on the human team. i also found the AI developers reporting before and after the event to be very disengenous.
P.S. My head started hurting too when trying to read the info.
I just wanted to pick out some baseline facts, and yes, Phantom is right, the more experience AI has against top pros, the faster it will learn.
Although if this is only level stacks and heads up, then, may I suggest, there is a way to go still, as full ring is much more demanding for a robot, than heads up.
It does show you how beautifully complex our game can be, when played at a high level, so much so, that computers can beat Chess, Go, but not our wonderful game!
If it was self learning that quickly I would have been hugely impressed, but it seems it was getting extra programming between sessions which suggest alot of its strategies are pre-programmed and not learned and the coders were plugging some leaks that the other players were exploiting.
I'v been watching this live and the bots are winning at 13bb/100 after 6k hands which is very scary!.... BUT And it's a big but.... Every hand is played with 200bb stack regardless of previous hands so it is only trying to solve at one stack size.... if this was played to go super deep the bot would more than likely get destroyed Posted by Itsover4u
out of interest why do you think humans gain an edge with stack depth?
One of the losers, D.Kim, is one of the better high-stakes HU players. The bot destroyed even him, which I think should suyggest to everyone that even the top players have an abundance of leaks that are just waiting to be exploited. Easy? No, but it at least shows how far even the best players are from being even close to perfect. One can try to memorise all of the strategies suggested to take by Snowie, or the plays suggested by Ben Sulsky, but the ability to instantly, or at least very quickly, adjust to a strategy & capitalise on any leak is the thing that wins the $$ at the end of the day. Most likely your strategy should be completely different vs 2 different players. Hyperexpoitation wins the money, not mass tabling
Well, it either shows that OR that the AI is just ridiculously amazing, you decide :~~~~~~
Comments
the previous bot, cepheus, was the same. it was reprogrammed between sessions to plug leaks and exploit the human team's exploitative plays. it even deployed different versions of the bot against different players on the human team. i also found the AI developers reporting before and after the event to be very disengenous.
Well, it either shows that OR that the AI is just ridiculously amazing, you decide :~~~~~~