These bots are supposed to be nearly impossible to beat. But I can understand part of kav's skepticism. How much of these bots strategies has a solid mathematical foundation, and how much is based on trial and error?
Notifications
Clear all
Fixed Limit
26
Posts
6
Users
0
Reactions
8,349
Views
I was looking for Polaris but I could'nt findt it anywhere. But I found Sartre, it seems to be much better than Neo, I played around 200 hands vs it, and I was slightly winning in the first 100 and then slightly losing int the rest. Not a huge sample size but I could at least determine that its not a fishbot. But its not a GTO bot, it plays like a regular 100% flp cbet never checks back, never donks, so its exploitable.
You can find it here:
https://www.cs.auckland.ac.nz/poker/
I wonder how could one win 16k$ vs it in 39 hands 
Originally posted by YohanN7
These bots are supposed to be nearly impossible to beat. But I can understand part of kav's skepticism. How much of these bots strategies has a solid mathematical foundation, and how much is based on trial and error?
Polaris is based on Years of Poker AI research from the University of Alberta so is based on very solid mathematical foundations and super advanced "shortcut" methods for reducing the size of the game (that plus the 2011 version had a decision tree with about 6 billion branches and even more data nodes)
The 2008 version of Polaris beat the best players in the world at that time and the 2011 version is 10x better in terms of how much a nemesis program can beat it for.
There's the weblinks to look up Polaris and some of the theses writeen by the people who designed it. It's definitely not been a trial and error job.
A learning AI like Neo or Sonia is a sort of "trial and error" situation becasue the bot learns and adpats, Polaris plays a set pre-programmed strategy, although hyper complex, which can only be changed by changing the rules of the game (which was done in 2008 to give Polaris "personalities")
this is very interesting, tho I don't think that a leaning AI is usable in poker because of its constant adaptation you just have to switch to level 2 and be always one step ahead of it. I'd really like to play polaris, for the sake of learning and challenge , but it seems that is was not published.
"Launch the polaris, the end does'nt scare us"
Unfortunately they didn't listen to enough Megadeth as it seems.
BTW meanwhile searching for polaris I stumbled upon a forum where bot devs discuss how to bypass the bot detection mechanisms of various widely known poker rooms. Lord knows whom you play at the tables these days...
When I got my "start-money" from PS I played at William Hill. The regulars were, as it turned out, obvious robots. But they were bad. They were simply too tight, and after some time I had their ranges exactly pinned down. But at PokerStars, I haven't seen any obvious robots, but who knows?
Originally posted by kavboj84
@Boomer2k10 :this is very interesting, tho I don't think that a leaning AI is usable in poker because of its constant adaptation you just have to switch to level 2 and be always one step ahead of it. I'd really like to play polaris, for the sake of learning and challenge , but it seems that is was not published.
"Launch the polaris, the end does'nt scare us"
Unfortunately they didn't listen to enough Megadeth as it seems.
BTW meanwhile searching for polaris I stumbled upon a forum where bot devs discuss how to bypass the bot detection mechanisms of various widely known poker rooms. Lord knows whom you play at the tables these days...
Well since humans are squishy and not GTO a learning AI should actually have more success than a GTO bot in that, there's no way a GTO bot will lose but the learning AI in theory should be able to absolutely crush beond recognition once it has a handle on the player base.
From what I recall (5- years ago now) Sonia absolutely destroyed human opponents of the time for a ridicuous winrate and even beat the 2008 version of Polaris by finding holes in it's GTO programming (although it is interesting to note that Sonia decided that almost mimicing Polaris's game was the optimal adaptation basically proving the concept the only way to fight GTO is with GTO itself) for about 0.18BB/100.
Now Sonia played incredibly wierdly in her default configuration (open limped pre, was hyper passive in some areas) and this was more that likely an adaptation to the population of the time
I believe Neo is similar in that its basic game is rather simplistic but after a certain number of hands you'll probably end up helpless against it, but that would be a lot of hands.