GTO Wizard AI
Really impressed with these advancements, but how exactly will GTO Wizard, or any poker site for that matter, be able to prevent people from using an online solver that fast as RTA?
So ive been reading through the responsesnin this thread and i find the discussion quite interesting. Im curious to learn more.
I have very very minimal coding and AI technological knowleg... but i had a thought that i wanted to share that may or may not be a possible solution...
Is there any way to use encription embedded within the source coding of a given online poker game so that the poker bots would not recognize what is happening on the board during live game play?
In other words can specific card values shown on the board be camoflauged or disguised so that the bot does not receive the correct card values, or perhaps receives no information at all, or even is caused to misread it entirely and not recognize it to be the board of a pomer game but something completely unrelated to poker.
Id love to hear feedback and of its even a possibility or not. Thanks
My best guess is no, that's impossible. If you can see the cards, then the bot should be able to capture the screen and get the exact same thing that you saw.
So ive been reading through the responsesnin this thread and i find the discussion quite interesting. Im curious to learn more.
I have very very minimal coding and AI technological knowleg... but i had a thought that i wanted to share that may or may not be a possible solution...
Is there any way to use encription embedded within the source coding of a given online poker game so that the poker bots would not recognize what is happening on t
Sites could use random funky fonts and/or different colors on the letters/numbers representing card rank to make the cards harder to read, but it’s not really a great idea for multiple reasons …
One being it would also be harder for humans to read. Two, it would just be a temporary obstacle … bot owners could just train their bots over time to be able to read the fonts.
So ive been reading through the responsesnin this thread and i find the discussion quite interesting. Im curious to learn more.
I have very very minimal coding and AI technological knowleg... but i had a thought that i wanted to share that may or may not be a possible solution...
Is there any way to use encription embedded within the source coding of a given online poker game so that the poker bots would not recognize what is happening on t
When the whole Houston Astros cheating scandal came to light, I remember thinking the counterstrategy would be to mix up the signals every now and then. Not in terms of the sequence/indicator but the actual number assigned to the pitch. So 1 becomes the slider, 2 = fastball, 3 = change, etc. You'd only need to do it occasionally for the system to falter. The first time a batter faced a heater while expecting a changeup, the confederates would know that a) they can't trust what they're seeing, and b) they've been caught in the act.
Similarly, I agree with the above that the AI would just get retrained for whatever change is implemented. But if the change could happen at random, unpredictable intervals, then the bot would not adjust accordingly in those spots. Like the batter getting fooled by an incorrectly stolen signal, it would be pretty funny to see some bot call off on the river with eight high because it "thinks" it has a straight or something.
Just look up the definition of standard deviation and do a sample problem for calculating it. Then you will understand.
The majority of hands we play in poker are just folds... so no money enters the pot. When you put in 3-bets, 4-bets, put in money on flop/turn/river, etc. all that will increase standard deviation. You can also just run whatever filter you want in your database and see the standard deviation. SRP are going to have lower ST
its a flawed argument..
Going less to showdown , can increase std deviation as much as always going to show down can , by always calling
on a perfect strat , easy example, if you can see your opponents hole cards , you would have a skyrocketing red line with the least std deviation possible and absolutely no downswings .
you're assuming that redline players fold less and bluff more , meaning they are actually deviating from EQ , meaning that they are making more mistakes , and that yeah will increase std DEV
Anyone have the issue of loading GTO wizard and it just being a black screen? I'm not sure what I'm supposed to do with that. I tried looking for a GTO wizard help thread on here, but can't seem to find one and hopefully someone here can help me out. Can't contact the GTO wizard people, because the only way it seems they contact people is through their discord which I don't have.
Just look up the definition of standard deviation and do a sample problem for calculating it. Then you will understand.
The majority of hands we play in poker are just folds... so no money enters the pot. When you put in 3-bets, 4-bets, put in money on flop/turn/river, etc. all that will increase standard deviation. You can also just run whatever filter you want in your database and see the standard deviation. SRP are going to have lower ST
This post got it. Seems like some people dont understand the formula though. You could also just explain it by saying the more money you put into the pot the more std dev will increase. I.e in HU you put in more money on avg as you say, and also 3b / 4b. When stacks increase std dev goes up. Std dev is a lot lower assuming you play short stacked.
Think the whole redline discussion is trivial though, but could be wrong about that. I think people read into it too much because most people overfold hence it increases your winrate to some extent. But also clearly when you play bigger pots on average your mistakes are clearly a lot more expensive. Comparing PLO vs NL is a good example i guess. Std dev is a lot higher in PLO despite winrates being twice as high on average. But you are also all in a lot with 60/40 equity.
This post got it. Seems like some people dont understand the formula though. You could also just explain it by saying the more money you put into the pot the more std dev will increase. I.e in HU you put in more money on avg as you say, and also 3b / 4b. When stacks increase std dev goes up. Std dev is a lot lower assuming you play short stacked.
Think the whole redline discussion is trivial though, but could be wrong about that. I think pe
just no..
this formula can be applied in poker , to get some data on the pools you're playing , but thats it ..
theres no correlation between money put in the pot and an increase in std deviation, without looking at other stats
HU std deviation is usually lower cause winrates are higher , again assuming that both players have the same winrate .. or that you're playing someone with a lower winrate that if he doubles up he might quit .. the bigger the edge the less std deviation you will have compared to 6max , so nothing to do with money put in the pot
std deviation is lower playing short stacked ? what .. mtts are purposely made to start deep to allow players to build a stack , and literally getting from an average stack of 200bbs to 30bbs after 5 to 6 hours, cause an mtt can't last 7 days.. clearly the shorter you are the more std deviation you will have ..
if you buy in for 20bbs in a cash game table , you're just lowering your winrate , you just lose less per hand .. thats one hand .. just cant interpret things the way its more convenient
PLO std deviation will naturally be higher on smaller samples , std deviation goes along with winrate .. so thats why you can't use this formula without introducing winrate.. over 1m hands , if you have a 10bb/100 winrate your std deviation will be lower than a 5bb/100 winrate at NLH .. one does not go without the other
comparing NLH with PLO , its not a good example , they are different games, but i'll give you that one , as im not 100% sure , the other arguments are very very easy to prove wrong
Edit: there's a variety of external factors that would need to be added to the formula, like hand sample, vpip , pfr, 3b , wwsf, winrate , stack size , average stack size etc etc etc to get close to make such assumptions
I cant find the roadmap anymore, whens the PLO slated for release?
https://www.youtube.com/watch?v=mi5F1-69...
Really impressed with these advancements, but how exactly will GTO Wizard, or any poker site for that matter, be able to prevent people from using an online solver that fast as RTA?
Some chess bots beat other chess bots.
I reckon a human will rule poker forever, becaUse GTO is inherently flawed. When it comes to timing,//tempo.
Has anyone been able to develop something custom that reads data straight from an active table an initiates a simulation? If not, that really seems like the next frontier of cheating capabilities, no?
If yes, presumably they'd keep that pretty close to the vest.
A cheater probably doesn't even need custom solutions for each spot. You can develop some sort of model that buckets similar textures and lines and just spits out some reasonable approximation. Surely with such a model you could quantify ex post how similar/dissimilar each output is and refine the model accordingly.
Thought experiment: if you were going to build a predictive model for identifying a cheater, what would be the most important variables to capture? I'm thinking
-Winrate
-Volatility (standard deviation of winrate) (honestly not sure if there's any predictive power with this one)
-Avg number of tables played at a time (if you still have to manually look up solutions you'd be pretty time crunched if playing a lot of tables)
-Anything to glean from average time to act? Either faster (because it's a computer doing the work) or slower (because they're constantly looking up solutions)
-What else?
Feel like comparing a credible sample of actions to Nash solutions and computing exploitability is not gonna work because it doesn't capture the intention to exploit so a cheater may very well be deviating quite a bit on purpose, as any competent person should wont to do it they think it's justifiable.
It's a two-sided thing: you'd want to know your own strategy but also opponent's response. So I feel like any anti-cheating model needs to incorporate population data to flag where people (assuming you've got data from a representative sample of non-cheaters) are most exploitable and specifically focus a lot on the potential cheater's behavior in those spots. But then you have this big issue of false positives: honest people may rightfully get pissed if they're penalized for being a sicko.
What about subjecting a potential cheater to some sort of monitored test? If they're flagged, they get randomly assigned a sufficiently large number of sample spots and, under timed supervision, they have to fully specify all ranges, strategies, and justify any deviations they'd make from Nash on the basis of their assumptions about opponents' behavior. If you can't actually do these computations from "first principles" but you're crushing then there might be a problem. Force people to "show your work".
I haven't really contemplated the effects of technology improvement on this stuff till now and tbh this reality sucks. This is a tough issue to tackle. And I feel like it should be taken as a given that nefarious people will make every attempt to exploit this.
Am I being too much of a tin foil hat conspiracy guy to think that there may be a shitton of people already using advanced cheating methods? Do online sites really have ANY mechanisms in place to stop this? Do they even have any incentive to stop it yet?
Oh ONE MORE POINT.
When you think about it, they don't really need to police everyone; just the people winning the most. If someone wants to set up a dream machine and half-use it to be a small winner, it's not ideal, but poker won't die. They could have just studied to become a small winner instead
Why not just let every single online poker player use an RTA standardized GTO bot with a "RTA" inbuilt and let 'em all fight?
A potential simple solution (?):
Mandatory screen recording software for all active monitors. And mandatory webcam. And lock down access to all web browsers and desktop solvers while the client is open?
Is this just too invasive to ever be an acceptable solution? Plus would sites have to invest in better infrastructure to process all the data?
Has anyone been able to develop something custom that reads data straight from an active table an initiates a simulation? If not, that really seems like the next frontier of cheating capabilities, no?
If yes, presumably they'd keep that pretty close to the vest.
A cheater probably doesn't even need custom solutions for each spot. You can develop some sort of model that buckets similar textures and lines and just spits out some reasonable appr
If everyone is using a model that Is equal it's fair game. If one is using a quantum generator of node/divergent//tree-route logic that goes helluva deeper then the competitor, then that's a huge problem. Think IBM deep blue versus chess computers now
A potential simple solution (?):
Mandatory screen recording software for all active monitors. And mandatory webcam.
Is this just too invasive to ever be an acceptable solution? Plus would sites have to invest in better infrastructure to process all the data?
You edited your post.
Screen recording is interesting. I think it poses more problems than solutions at current;;
Time delays, deep fakes, and collusion
Invasiveness is another issue. Lots of sites have tried and failed.
The average winning non-cheating player seems to base most of his game on heuristics based on things like formation, action, texture etc, and the improvements come from expanding and refining those heuristics. Add MDA for some river exploits and you have it. If a bot can just be like that, and execute a very good human strategy without ever making a mistake (inside the human strategy) or tilting, I think he will be winner. Not perfect at all, but perfect is the enemy of the good (good=stealth).
Is everything I said above implementable by any decent programmer? Absolutely.
The infrastructure to execute this would be the main problem from my pov.
EggsMcBluffin:
- Winrate first.
- If all of this end up being deployed as bots, I assume it would be deployed with tons of accounts playing the same way, so I guess a site would have to constantly watch for players playing the same.
- Someone playing big volume, picking random 100k, 200k samples and scrutinizing the stats just like you would do to catch a full botfarm, as I don't think any human would be able to execute the exact same strategy he did in 100k hands, over the next 100k hands. Including showing the same leaks.
- + anything else others said or may say itt 😀
The average winning non-cheating player seems to base most of his game on heuristics based on things like formation, action, texture etc, and the improvements come from expanding and refining those heuristics. Add MDA for some river exploits and you have it. If a bot can just be like that, and execute a very good human strategy without ever making a mistake (inside the human strategy) or tilting, I think he will be winner. Not perfect at al
Bloom; first paragraphs true
As for;;;
- how do some bots make more money deviating from GTO for exploits from data affect winning players ~potential win rates, IF that's in their programming
- no bots, fish whales and sharks and shrimp
- all bots, the well dries up
Also, worth to think;;
If bot farms are targeted and one true player gets banned, is that worse than allowing break even slightly ++ bots from winning????
Any bot running at even close to 70%+ optimal should be sniffed out ez
I believe fish gonna either lose less from bots or worst case the same than vs regs. Bots are bad for regs in my opinion, and most of the complaining is just self interest.
Unfortunately, lots of fish fear bots, rta and cheating, they fear or used to fear huds lol. With no fish, regs will slowly disappear, and no players=no games.
My 2 cents
I believe fish gonna either lose less from bots or worst case the same than vs regs. Bots are bad for regs in my opinion, and most of the complaining is just self interest.
Unfortunately, lots of fish fear bots, rta and cheating, they fear or used to fear huds lol. With no fish, regs will slowly disappear, and no players=no games.
My 2 cents
They play vs fish way better than the average reg
I believe fish gonna either lose less from bots or worst case the same than vs regs. Bots are bad for regs in my opinion, and most of the complaining is just self interest.
Unfortunately, lots of fish fear bots, rta and cheating, they fear or used to fear huds lol. With no fish, regs will slowly disappear, and no players=no games.
My 2 cents
Live poker boom incoming.
My 2cents online poker??
Old ''regs'' will become low losing recreational
Some bots will eak out ~2bb/100
Some meta-sharks will have 5bb win rate
My bad, fish has very predictable leaks that any bot programmed with mda data and the adjustments a good reg would make will just destroy. It’s gonna be a massacre lol
A bot would never 2nd guess a close bluff catch for stacks vs a fish after losing 10 times in a row because he thinks that fish is different than the average fish or whatever.
just no..
this formula can be applied in poker , to get some data on the pools you're playing , but thats it ..
theres no correlation between money put in the pot and an increase in std deviation, without looking at other stats
HU std deviation is usually lower cause winrates are higher , again assuming that both players have the same winrate .. or that you're playing someone with a lower winrate that if he doubles up he might quit .. the bi
Nice attempt to patronize. I'll give you a hint and I'll leave it at that. The unit you measure std deviation in poker is bb/100. Variance in shortstacking poker (and mtts) is lower in terms of bb/100 but higher relatively to your stack. You will get stacked more often since you are all in more often which is good for the speed of the tournament. But the swings and the volatility of them will be far lower the shorter you are. High risk = high reward , small risk = small reward.
