"The Russian Bot Army That Conquered Online Poker" (Bloomberg)
This is posted in the bot thread but I think it deserves a separate thread as its a good, well researched piece that is
Please don't derail the derail.
These comments were made almost six months ago by Noam Brown so they aren't a reaction to this article but they are obviously on rail for this thread.
[QUOTE=Noam Brown]I can say with confidence that AIs are superhuman at poker. Even if you have issues with the fact that Pluribus was only 6max and 100bb stacks, Pluribus was 5 years ago. The technology has continued to develop and the latest poker bots are superhuman at basically all variants. The only thing they are bad at is exploiting opponent weaknesses, meaning they might not make as much money off of weak players as a pro would. But after Cicero, the techniques to enable that exist.[/QUOTE]
Q : What do you mean by the latest poker bots, are there any that are multi-player and not limited to fixed starting stack sizes?
[QUOTE=Noam Brown]Yes, there are poker bots that are multi-player and not limited to fixed starting stacks. There just aren't papers written on them. The best ones I know of are based on ReBeL but extended to multiplayer. It's not as hard as it might seem because usually the game is heads-up postflop, so you mostly only need a good multiway strategy for the preflop where ranges are over 169 hands rather than 1326.[/QUOTE]
Pretty crazy find that haven't seen posted anywhere in the other threads before now. Its a zoom call in English with the developers giving a full overview of the product, how it works etc at the time in late 2021 to current customers and potential ones, with a Q&A afterwards!
Pretty crazy find that haven't seen posted anywhere in the other threads before now. Its a zoom call in English with the developers giving a full overview of the product, how it works etc at the time in late 2021 to current customers and potential ones, with a Q&A afterwards!
Holy ****, nice find. They have even built in realtime card sharing, sick.
16:50: why robot folds strong hands?
"our robot knows [..] the cards of other players with NZT in the hand and applies card blocking effects. that is why we recommend sitting with 2 or 3 accounts with NZT at the table. the robot will have more information to make the best decision."
Of course they did, why am i even surprised?
My thought (having started an AI startup 6 ish years ago).
- It's really easy to make an AI Poker Bot. A +EV bot can be completed with 1-2 days of work.
- You'd need time to test and figure out how to detection and make it more user friendly, but a working +EV prototype can be made quite easily.
(most of the codes can be found online / asked ChatGPt to write it) - If websites use bots (scary thought) to balance the playing fields towards the weaker players, it would be nearly impossible for players to detect as house can make unlimited new accounts. It would not leave enough data behind for players to detect.
- I do not think that this would be done at big sites as other alternatives exist
e.g. GGCare Flipout - losing players get easier entry registration to GGCare Flipout
- losing players get more favorable seats in GGCare Flipout
- losing players get wayyy higher rakeback
I am NOT at all pessimistic about online poker's future- The bots /RTA will get worse before it gets better
- I think an online poker room will eventually advertise their room as "bots free" and requires camera / human interaction - when players move to new rooms, other poker rooms will also have to follow
- Possibly online poker room will start attacking each other by ousting others' bots
- It's actually quite hard to make an AI video / image to mimic human at real time | solutions are not difficult to implement
- e.g. players are required to have mouse / pen / eraser - and a random prompt shows up and ask user to hold up one of the items
- e.g. players are required to have mouse / pen / eraser - and a random prompt shows up and ask user to hold up one of the items
- I do not think that this would be done at big sites as other alternatives exist
Pretty crazy find that haven't seen posted anywhere in the other threads before now. Its a zoom call in English with the developers giving a full overview of the product, how it works etc at the time in late 2021 to current customers and potential ones, with a Q&A afterwards!
From the presentation you can really tell that they're structured very much like any other legitimate software company, with a meeting moderator and a product manager and everything.
also: I'm baffled by the underwhelming response to this thread. Maybe it's too confronting and people just don't want to hear it? I personally find it very fascinating to get an insight into the world of BFC but on the other hand I don't depend on an income from playing online poker.
Starts in 5
also: I'm baffled by the underwhelming response to this thread. Maybe it's too confronting and people just don't want to hear it? I personally find it very fascinating to get an insight into the world of BFC but on the other hand I don't depend on an income from playing online poker.
Which parts of the article were eye-opening? I thought bots from Eastern Europe shitting up online poker was just common knowledge by this point. Their scale and organization shouldn’t be a surprise. Botting in online poker is insignificant compared to the cheating industry for online games.
Holy ****, nice find. They have even built in realtime card sharing, sick.
16:50: why robot folds strong hands?
Of course they did, why am i even surprised?
Crazy when you consider this video is from around 4 years ago. I caught him saying in the video that in the player block they are taking in 253 parameters to give an output just for that block. It ties with what Pablo was saying earlier in the thread about fighting bots ever since 2010. These guys obviously have this system that they somehow built with MDA, player stats, conditional logic and skilled poker players all together to form some sort of the system that was playing poker before calculating the NE appeared. It seems like they are trying to take some of the bits from the different systems and create a bot that at baseline plays GTO while using the other system they developed over many years to also exploit players.
This would have been roughly around the time Sauce
on the Brandon Adam podcast. The bots got better but the sites got better at detecting them. I think the bots will get to a point again where they can play without detection but I would be very surprised if even on the sites with the best security that RTA isn't a big problem. You can't really bot at scale without regs at mid/high stakes seeing something is up. Humans cheating with machines I would say is still the biggest issue in many games. There is loads of private groups using this software. I know someone said above you can build something in a few days, you don't even need to, the screen detection is done, find some API's and away you go. Most of the private solutions I would say are people reselling this exact setup in some capacity.
Good rundown of the public software market here, I am not sure exactly when Ruse was developed but I first seen Rabichow reviewing hands from it in late 23, shortly after which it was acquired by GTOWiz. These AI solvers are in the wild in private, 100%. Little bit of network/computer knowledge you are untraceable. I don't care what kind of 'trust me bro' crap the site is feeding people.
Matt mentioned something on the pod that I wasn't aware of in regards to WPT Global. They classify and skill match players on the site already. Imagine if they are using software from Deeplay to do all of this? 
How do WPT Global determine skill?
High skill players are labeled by a proprietary ML (machine learning) algorithm that utilises strategy & behaviour features to calculate skill classification. The model is trained by using gameplay data from top winners on our site with the goal to identify long term winners at an early stage
How early is the skill classification applied?
The model produces a prediction in the first 300-500 hands with ~90% long term accuracy. The system continues to evaluate players as they play more hands to adjust the label accordingly. Precision increases as the player’s sample size increases
How is skill level determined at each stake?
- The system classifies skill levels separately at each stake to take into consideration different skill & play styles in various player pools. This also prevents players from manipulating their play style at lower stakes to gain access to higher stakes games.
- If a player is labeled PRO/high skill at any stake, they will be automatically labeled high skill at all lower stakes
- If a player is labeled Pro/high skill at any stake, they will not automatically be labeled high skill at higher stakes
It is possible for a player to be labeled PRO at at a given stake but labeled non-pro in stakes/games above their PRO label
How is the skill label applied to the seating mechanism?
Non-Pro View
Recreational players can view all tables in the lobby and have access to all available seats
PRO view
- Configurations are set which cap the number of PRO players that can join the table
- PRO restrictions vary depending on the makeup of the player pool and are subject to change at our discretion
- When PRO seats are full/occupied at a given table, that table disappears from the lobby for other PRO’s
Software is provided by a company called A5 Labs, I've no idea if they have links to Deeplay.
This is the whitepaper for their vision of online poker. It is well polished, sounds good on the surface but ultimately I think its a polished turd which is just more dystopian nonsense on creating balanced games in which ultimately the winner is the rake -
Had had a quick look at the WPT thread on here for the site, it seems like a dumpster fire of a site. They seem to have had alleged issues related to a recent superuser as a cherry on top.
The overall online poker landscape is absolutely crazy.
The bot situation on GG Poker is out of control. In cash games, particularly in Rush & Cash across all limits from 200 to 2, there are a large number of bots—around 10-30% of the player pools. These bots are playing in a very obvious non-human manner, and GG Poker does nothing even after these players are reported. For instance, these bots use the snap cashout button or the snap emoji button, which indicates that the bot users don't even bother to hide them.
Awhile ago people had thought Party Poker had started grouping players by skill level and then stopped after the bad press.
Awhile ago people had thought Party Poker had started grouping players by skill level and then stopped after the bad press.
Party tried segregated pools based on player ability. I believe they ring fenced certain types of players and then they could only player with each other. This is different in that they are trying to balance the tables with a mix of different skill levels.
Edit - Obviously this is what they are saying, how it works in practice I suspect is quite different.
Party tried segregated pools based on player ability. I believe they ring fenced certain types of players and then they could only player with each other. This is different in that they are trying to balance the tables with a mix of different skill levels.
Edit - Obviously this is what they are saying, how it works in practice I suspect is quite different.
Did Party stop doing that, or still going on now? I remember hearing about it when it was new / being considered. Seems like a dumb idea.
There are always going to be winners even if you divide pools by skill levels. The only difference is some regs who once were winning won't be able to anymore, and they will quit. A few of the "better" fish will win against other fish since someone has to win. This will cause the skill levels to be adjusted again, the process will repeat, until all that is left are the fishiest players ... traffic will take a nose dive since in a normal environment (without separating pools), regs are the ones starting tables in most cases.
Did Party stop doing that, or still going on now? I remember hearing about it when it was new / being considered. Seems like a dumb idea.
There are always going to be winners even if you divide pools by skill levels. The only difference is some regs who once were winning won't be able to anymore, and they will quit. A few of the "better" fish will win against other fish since someone has to win. This will cause the skill levels to be adjusted again, the process will repeat, until all that is lef
The party thing was way back many years ago. They were purely using winrate which is very problematic in itself because of the variance in poker. It was defiantly removed back then as there was huge outcry from the community because all of this stuff was extremely alien back then. Given all we know currently, who knows what each individual site is actually doing these days.
The difference here is that WPT claim to have some kind of prediction model that within 300-500 hands can classify the expected value of a player's win rate with 90% long term accuracy. They use it to award the player a skill rating. This then determines the tables which you can see in the pool. They say that if tables already have two or more players above a certain skill rating that table isn't visible to other 'sharks. This is dynamic in nature, so it is constantly getting updated and being used to balance the ecosystem. The overall makeup of the pool at any particular time I presume also goes into what the classifications end up being at any particular time. There are going to be times when the highly skilled player can't see any tables because they are all full of 'sharks'.
Add in a few highly skilled house liquidity bots that get added at the top of the queue as the allotted 'sharks' on particularly soft tables and you have the perfect ecosystem for the site.
As a non-poker player (Not even a rec) my opinion here might not be worth much here but here it is anyways.
The actual solution for farms on poker sites is pretty straightforward. After all sports betting has a similar issue but of course the websites are so much more heavily invested in not losing money because the money that's being lost would be their own.
(Those who understand sports betting know that it's not incredibly difficult to get a 56% win rate but that sports betting sites make money by not allowing those players to play for long)
The way it works in the US is every account needs a ssc that matches their name. They need to upload identity that matches the SSC. And all withdrawals go to an account name that matches the other two. Not sure what the equivalent identity number is another countries but it could be matched. Sure you can get your mother or your brother but very soon you run out of accounts. Plus people don't love large amounts of money coming through their account so it's been a challenge anyways
People still find ways around but it's a very very small minority.
In fact it's a shock to me that the big sites don't run like this already. I guess there's tax implications to some people.
Of course people cheating with RTA is a much bigger problem... But if sites were to ban you if you get caught and it's hard to get anothe account..
This is a story that fascinates me, similar to alpha zero when it beat stockfish it portends greatly and led to me being much less surprised by chat GPT and the AI explosion years later.
Here we should have an idea of what the stock market will look like 20 years from now where human trading will more or less be silly. If that's not already the case. Other implications as well.
But to me the moral implications hit differently as does the poker industry future.
I don't feel the evil element of these farms in the same way sure it's a step further but poker was always about sharks eating fish. Sure ethically there was nothing wrong, nobody forces anyone to play poker. But it's somewhat of a moral ambiguity when you know the only way you can make money is by fooling another person into thinking they have chance when statistically over the long run they have to be losing an order for you to profit.
Especially post GTO, because sure human error is involved and anything can happen on a single given hand. But by and large you're pitting a pool of players against computer solutions (via human memory) and slowly draining their funds. I always laugh when I hear people talk about growing the game like it's chess or basketball. Do they mean improving everybody's skill? Not quiet. It's about getting more fish to feed on
So you had a pool of sharks slowly feeding on fish using computer solutions to mathematically give them a huge edge. Along comes a bigger shark and finds a method to directly employ these computer solutions and feed on these other sharks. It's like a shark that learned to use a spear gun lol. Is there a definite difference ethically? Of course. Morally however?
The future of poker online may be dead. But is that a tragedy?
Through most of human history card games were largely a recreational activity with occasional hustlers who made a living sweeping through these recreational games. Or a losing activity they do when they go to their occasional vacation at a casino.
If we return to the state who loses out? Not the Recs, they were losing already. Not society, a slight brain drain that poker puts on our population surely doesn't help anybody. Taking above average intelligent people with good work ethic and locking them into a game where they just take other people's money is not productive. Maybe society's loss of entertainment? But how many non players are actually watching skilled poker? Professional poker has long lost the entertainment factor. It's rec poker that people find entertaining if at all.
The sites lose out and the pros. You can make an argument that a large portion of pro players would have happier lives If they did something else. Surely we all know pro players whose lives would have been much happier without. Or even worse, (I had a friend who was a grinder who killed himself -was quite a good player too he is unfortunately not a unique story)
Good conversation below hosted by Patrick Howard on RTA/Bots//Security/State of online with Alex Scott, (President of WPT Global) and John Andress (Former poker player who is part of A5 labs mentioned above). Everyone will judge for themselves in relation to what WPT are doing but for sure I think the transparency around it all should be applauded. Its great to see these guys out in public addressing the concerns and tackling these really hard topics.
I find it kind of incredulous that the likes of Poker Stars have hardly addressed these issues at all in public. If anything it seems like such a missed opportunity to try reclaim some market share given your main thing is that you are the most secure site. They probably have a policy of not commenting on this stuff but that horse seems to have bolted. This is a huge mainstream article and even people who don't play poker were asking me about this story.
Good conversation below hosted by Patrick Howard on RTA/Bots//Security/State of online with Alex Scott, (President of WPT Global) and John Andress (Former poker player who is part of A5 labs mentioned above). Everyone will judge for themselves in relation to what WPT are doing but for sure I think the transparency around it all should be applauded. Its great to see these guys out in public addressing the concerns and tackling these really hard topics.
I find it kind of incredulous that the like
I appreciate your words and sharing the video. I'm also, I think, the only person representing a poker site who is quoted by name in the original Bloomberg article.
I know a lot of people, especially on 2+2, will look at what we're doing with a lot of scepticism. I hope that by being transparent about what we're doing and where the challenges are, we will earn the trust that we need to be successful.
I've never met a team that understands the bot problem better than the group at A5 Labs and I am optimistic that we will succeed, but I also suspect that the industry as a whole, and particularly smaller sites who don't have the money to invest in this, will be harmed very badly before this fight is over.
I'm speaking at a conference in Dublin in a couple of weeks time, alongside Jonathan Raab, and we'll try to publish some of the discussion for those who are interested.
If anyone's interested - here are the details of the talk Alex mentioned at EDGE that will focus on the topics in this thread.
Wednesday 30th October
3:30 pm - 4:30 pm GMT (11:30 AM - 12:30 PM Eastern Time (ET))
Rise of the Machines: Can Human Players Survive the AI Bot Invasion?
Terje Bremseth (CEO and Founder - New Wave Poker)
Jonathan Raab (Gaming Industry Pioneer)
Alex Scott (President - WPT Global)
Kris Galloway (Head of iGaming Product - Sumsub)
- There should be a stream too, I'll link it here once it's confirmed.
From the Deeplay website:
Deeplay mission is to provide a comfortable environment for gamers. Our robots employ different strategies to maintain in-game balance. So, non-professional players lose less, enjoy game more and keep staying in the game longer.
I fully believe GGnetwork partners with them to balance their games to ensure the pros don't skin the fish too fast too quickly. As poker players, we are seasoned in pattern recognition & the longer we continue seeing the same patterns the more we believe their existence.
I play ~40k hands monthly on GG and the daily pattern is the same for me. I will grind 1k-2k hands, slowly stacking wins where my daily graph looks like a slow climb. Once I've won several buyins on the day, the site determines that I've won enough for the day and need to give some of it back. It will destroy me with a bad beat like AA v AKss or a 1-2 outer. Once that first bad beat happens, that's the sign to stop for the day, otherwise the setups will follow you like death in Final Destination.
If you are smart enough to adjust your play to be more passive & get away from several setups, the cooler setups will continue to worsen until you're dealt something like set over set or AAvKK with the K spiking the flop because it will force you to stick your funds in the pot in some way without your V drawing dead. The problem even with stopping for the day after that first beat is the algo will be waiting for you the next time you login. It'S jUsT vArIaNcE On the days you stop early, you can expect the next day will be miserable until those funds are returned to the eco-system with a small bit of interest.
Call me a rigtard, I don't care, but I can see the pattern because it happens nearly every single day if you play enough. I believe that's part of the reason GG can be so good with their reward system because they know the algo's will help the fish with coolers & 1-2 outers to stay in the RNG balanced eco-system where it's a "comfortable environment where the non-professional players lose less to keep them in the game longer". Then you need to distract & placate the pros (who are getting bullshtted at a consistent rate all of a sudden) with additional rewards & at a higher frequency so the dopamine does its job & brings us back the next day. They merged casino psychology into a poker room.
During the weeks I'm losing bad to bullsht beats and losing 80%+ hands nonstop, those are the weeks I cash in the GGcares multiple times usually with one of them being a deeper run. Also I seem to hit the top rake spin reward too. During the weeks when I'm even a slight winner on the week, its a dry spell and I will never cash in GGcares and I will get the lowest rake spin reward every time. There's too many patterns that I see daily to ignore.
At the end of the day, they want everyone, pros & casuals, breaking as close to even as possible while they continue to scoop in millions in rake.
From the Deeplay website:
I fully believe GGnetwork partners with them to balance their games to ensure the pros don't skin the fish too fast too quickly. As poker players, we are seasoned in pattern recognition & the longer we continue seeing the same patterns the more we believe their existence.
I play ~40k hands monthly on GG and the daily pattern is the same for me. I will grind 1k-2k hands, slowly stacking wins where my daily graph looks like a slow climb. Once I've won several buyins on the
LMAO.
Not sure if serious.
