i think people who cheat do well at online poker ,if you're sat at a table where a few of the players are colluding then you're at a massive disadvantage , I think that's not so hard to do online
@@filippocastelli42 I could try an organic GAN by abducting Nocoldiz and Frederick Frankenstein, tied 'em up and make them work on the same video until it comes out perfect. Don't know if that would be legal, though :)
This paper has been out for a month, I'm sure some smart people are quitely using it to their advantage already, likewise to when Snowden dropped the NSA info, till today hackers are using those vulnerabilities.
Funnily enough though, it doesn't. The AI only picks what it wants to do, not WHEN it wants to do it. So no really long wait to emphasize a bluff and make humans more antsy. No immediate raise to project strength (because everythin appears to be immediate. A good play can use timings as well.
Yeah, I would argue that if you're not a human (or an alien with readable emotions) you're not actually playing Poker. The biggest reason for using money in Poker is to influence people emotions to make them more readable. Any other game with real stakes, statistics, and bluffing will play largely the same as Poker.
Always great to see papers with accompanying blog posts that make them accessible, while still explaining the broad ideas. Highly recommend anyone reads it if they haven't already.
So there's a lot of misinformation floating around about this A.I. #1 the sample size was only 10,000 hands. Which to some of you may seem like a lot but to actual professional poker players they consider a sample size of at least 100,000 hands to see if you're a true winning player and to negate most variance. #2 the bot had an extreme difficulty making computations with varying stack sizes and its performance worsened the more the stack sizes varied. So, while impressive I would not consider it better than humans until I've seen more data
There seems to always be misinformation revolving around AI in general. Probably because people want to believe we are living in a science fiction movie were the evil machines take over the world
I agree 10,000 hands is a joke. No way it accounts for variance. 100,000 isn't even enough. Even at most optimistic estimates, 300,000 is the minimum but most poker players will demand much more sample sizes.
Pfft it's better. The higher the win rate, the less hands you need, and it was crushing the game. If it's not quite as good as stack sizes diverge, it can just quit and join another game.
Okay, I'm not a machine learning expert, but I don't understand where the "secret sauce" is in this paper. What, if anything, was stopping people from using this approach a long time ago? The paper mentions that the training algorithm isn't very different from ones used by previous poker AIs, and it didn't take a huge amount of money or specialized hardware (about $144 on commercial cloud services!) to train, or even to run (two high-end CPUs worth about $4,000). So why did nobody reach nearly this level of performance before? I feel like I must be missing something important here. Usually, even if it's not clear *why* it works, I can at least get a sense of *what* was different. Here I just have no idea, but clearly they did something significantly better than anyone before.
You raise a good point! The success of the Pluribus project was not due to any major breakthroughs in machine learning algorithms or hardware, but rather to the careful integration of existing techniques and the development of a novel approach to game-playing. One of the key innovations of Pluribus is its use of "nested subgame solving," a technique that allows the AI to reason about the game at multiple levels of abstraction and to make decisions based on the expected outcomes of subgames. This technique enables Pluribus to handle the complex strategic interactions between multiple players in a way that previous poker AIs could not. Another important factor in the success of Pluribus is its ability to learn from its own gameplay and to adapt its strategies in real-time. This allows Pluribus to adjust its play style to exploit the weaknesses of its opponents and to avoid being exploited in turn. This kind of adaptive learning is a relatively recent development in machine learning, and it requires sophisticated algorithms and large amounts of data to work effectively. Finally, the Pluribus project benefited from advances in cloud computing, which made it possible to train and run the AI using relatively inexpensive hardware and commercial cloud services. This reduced the cost and complexity of the project and made it more accessible to researchers and developers outside of large corporations or government agencies. Overall, while the Pluribus project did not introduce any radical new techniques or technologies, it represents a significant step forward in the field of AI game-playing and demonstrates the power of combining existing techniques in innovative ways.
In "Example Hand 2", I don't think that huge check-raise on the River (4.7k raise in a 2k pot) with that hand was a good play. P6's call in response was also questionable, his hand was only beating bluffs at that point.
AI had good reason to think the human had nothing but a weaker pair-hand from the flop. Human player probably knew from many previous hands that the AI was unusually willing to make big bluffs in such situations (compared to most humans), and felt compelled to see if the annoying AI was trying to force him off a hand "yet again."
That was a very weird play by the AI. Most opponents would fold, I'm pretty sure I would. It looked way too much like it was trying to show weakness and then pretending to bluff.
Lol that 'trap' with the QJ was a disaster, with that river raise you only get called by better hands. On top of that, humans trap their good hands all the time lol
@@lutherschultz4725 I disagree, as the human, you never get three streetz of value from worse hands there, so you either check back turn or check back river to control the pot
@@DaverendeDodo imo it's better to check back river than check back turn, because you let the opponent draw and limit your range on the turn when there is still action on the river. Unless if you include enough check backs on the turn with your tptk+ hands I think it's a bad play.
The players have fixed stack size in this paper. If you consider real poker environment, player's stack can varies from 1bb to 300bb. Sure the most popular stack size is 100bb, but in real life there are can be different stack sizes, which can change strategy because you play with different potential pod odds
Okay pluribus is pretty good although he played in a 6max format against players who weren't experts in 6max cash but in mtt's (tournament format) Which is pretty important. Also I heard that after the end of every hand the stack sizes were reset to 100 big blinds. That means the bot doesn't know how to play short stacked or deep stacked poker only the standard 100bb. Also some of the hands played by pluribus were analysed in a gto (game theory optimal) solver and it basically showed that even the AI wasn't playing close to a perfect strategy. I'd like to see a rematch but this time against the top 6 max players and without resetting the stacks. Only then you could say that AI can beat humans.
It played against Linus Loeliger (believed by many to be the best 6-max cash player at least at the time) and Nick Petrangelo (probably knows enough to beat some high stakes cash online even though he's a tourney player, arguably one of the best). I don't really know the other names in the list of players it played against, they're pros from slightly before the heavy GTO era e.g. Dong Kim I believe. It wouldn't make any sense to play SnGs, tourneys or varying stack sizes to test the bot because then you would have extremely messy and meaningless results. Sng/Tourney is much higher variance and humans don't want to play for that long, and measuring its performance at 20bb, 40bb, 60bb, 80bb, 100bb, 120bb, etc would just be silly and subject to variance. Even in this low variance format, the results are noisy. Pluribus plays preflop ranges very close to accurate preflop solves. If it deviates from what a solver says post-flop it probably does so in a way that doesn't blunder EV - I'd like to see an example of that if so.
It was a genius play if you think about it. The back-check on the turn of P6 was actually really bad because the AI had to know that KQ or AQ is going to bet 3 streets. So either the AI gets a massive amount of value for having the higher kicker or it wins the pot anyway, because no weaker hand than QT is going to call the reraise on the river. I think its quite profitable in the long run. If P6 would have c-betted the turn. The river would have been a check-check and P6 wouldnt have lost so much chips.
"This program was brought to you by Weights, and Biases" That reminds me of Sesame Street where they'd list numbers and letters of the alphabet as sponsors for the episode.
I love AI playing games, it's amazing seeing it conquer game after game. It'd be cool to see machine learning applied to video games like FTL or Slay the Spire
This is the death of poker ;) Nobody will play this once these AI's are around to dominate the game both online and live where people will cheat with earplugs ;)
Perhaps as there is so much money involved. Though I still find it unlikely as AI did not kill other games such as chess and go. We will just need to develop even better rules to ensure there is no cheating
MrFram Competitive online chess, go, etc aren’t dead. Why is this different? We have systems in place to detect people using bots to inform their decisions...
The hand that was shown can not be evaluated out of context. If a beginner showed me how he had won this hand and asked me for an evaluation, I would tell him that he overplayed his cards and was lucky to get called by pretty much the only worse hand that might pay him off. However, now that we know that the hand was played by a strong bot, that has established a table image of being a frequent bluffer, we can reevaluate the river check/raise as a brilliant merging of ranges. Knowing that it is capable of doing this, the humans can no longer consider a river check/raise just a very polarised bet that either means a total bluff or a super strong hand, like three of a kind, but now they also have to consider that it may be a medium strength hand, where they have to decide if their kicker is good enough. This is scary. I am happy that I stopped playing online poker for money a long time ago. This is the official death blow to online poker, since it is now impossible to know if you are playing against a complete noob or a superhuman AI before you have lost so many hands that it is clear that it is not some random luck box.
I was completely with you until "This is the official death blow to online poker". It's going to be dangerous for pros playing each other, but most online players are still fish who neither are aware of Pluribus nor know how to use it.
The highest levels of poker are usually face to face, where quick reads and expressive control play a big role. I wonder if there will be an AI that can make a heads up read and tell if a person is bluffing or has a strong hand. I think that could arguably be more powerful at a table, and scary!
You'd want a spy pen to bluetooth video to your phone, which would then upload it to a server that did the actual processing. An earbud could then tell you what to do. You'd still need a pokerface to bluff though :P
I'm blown away by the quality here. I recently enjoyed a similar book, and it was top-notch. "Game Theory and the Pursuit of Algorithmic Fairness" by Jack Frostwell
I don't like your explaination for the A.I play... she can't know that her kicker is better as the player would, assuming he has a queen, raise pre-flop with Q-10 to AQ (so again assuming he has exactly one qeen it only beats 1 out of 3 hands he can have). The check post-flop from the A.I out of position is pretty standard, so is the c-bet from P6 and the call with top pair and possible straight draws. The A.I. checking that turn is again pretty standard, especially since the villain was the aggressor pre-flop and that you have top pair (so you want to induce a bet if you're opponent thinks he can bluff you of the hand. The check back from P6 makes sense since he has no more draws with the turn and the possibility that the A.I. is slow playing a better hand (AQ/QQ/KK/AA/99/77... although I assume the A.I would have 3-bet AQ/QQ/KK pre-flop), plus the fact that he wont get 3 streets of value from a worst hand. Now the check raise from the A.I. on the turn is really weird, especially considering the sizing of the raise, you would expect it to have either be a bluff or a monster, not a top pair 3rd kicker. It's very unlikely that the A.I.'s hand improved post-flop (the only hands that would be pocket deuces or threes. So if assuming the A.I. 3-bet AQ/AK/QQ/KK and maybe some smaller pairs pre-flop, the only hands in its very wide big-blind calling range that beat/tie you are AA/99/77/22/33/KQ/QJ/Q10/97(suited) and that's about it. As I said, I wouldn't expect such a raise from Q10 to KQ since you can easily be out-kickered by P6 so in my head the A.I has either one of the monsters I listed or one of the many more bluffs it could have check-called up to the river. For that reason (and unless we know the A.I. can slow playing hands like AQ/QQ/KK pre-flop out of position) the call makes some sense although you only beat bluffs. It's really that huge raise from the A.I. that's weird and probably a bad play, has it will have most worst hands (exept some Q10) folding but most better hands (exept some QJ/KQ) calling (and monsters raising all-in)
As well as that I know that if I had a q 10 on the first hand and someone that was playing passively made a big raise I'd either re raise them if I was planning on buying in again or if not play it safe and cut my loses
You can't get any conclusion out of only one hand. According to the briefs "interviews" of the players who played against it, it was fond of bluffing, and was especially good on the river. There is a game history that is not shown here (because it's a 5mn video talking about a paper and not a 1 hour video analyzing it's poker play), and it likely contributes to why this hand worked hat well. That being said, I agree that this hand doesn't seem anything special to me. Also, you can't rationalize how an AI plays using our human logic, it just doesn't think like that. And this is what is so interesting about it. In chess for example, since alphazero and leelazero games started to show off, some top-level way of playing and evaluating position has started to change a little bit, because the AI gave new ideas as to how to understand and evaluate positions. This is the same here, if you stay stuck on the human way of understanding poker, you won't get the chance to improve with what the AI shows. But once again, we can't extract anything meaningful out of only one hand, we agree on that.
I think the huge raise is a bad play only if you think that most people will fold to that. Maybe the AI has learned (and is able to recall) that 20% of the cases, people will call his raise and therefore the overall value made on the hand will be higher compared to making a smaller raise (that is called more often, but not often enough to make up for the difference in total money won). So that is why the river play seems crucial (at least according to this example hand).
@@Sanzarc The problem isn't the A.I. trying to get value but the hand it's trying to get value with, as it can be beat easily and that most of the time only worst hands will fold
River check-raise seems a little bit thin for value but I suppose the button's range is somewhat capped after the turn check back. I just don't see a lot of worse hands that will call the raise (other than the exact QT button had). Maybe it's a bit of a merge play as you would fold out some KQ and maybe AQ while still getting value from a couple of worse Qx hands? I'd love to see the analysis because one hand where you have your opponent pipped doesn't show much.
the problem with this scenario is that whatever test subjects they used are likely not playing at the level they would be if the stakes were high (bracelet and purse)
Please tell us who are the two wsop main event players the AI played against. Surprising to see that none of those players admitted anywhere that they actually played this game against AI
How do we know this isn't a trick by online Poker companies to try and get us to play their online Casinos? And then the bot makes us lose all our money????? 😱😱😱
If you play on any legitimate site I wouldn’t worry about it. Besides, any readily available ‘bot’ that can play for you undetected on a site these days would actually be losing money
Wow, that example hand was nuts. Mainly because any good poker player knows that you generally go aggressive there and raise. Not raising until the end is thought to be super risky. Generally, you just want to win your pot ASAP. Very interesting stuff!
FYI in some of the examples shown both flop and river play by human were bets not raises. However I am sure the bot knows this after all only humans make mistakes
Note a possible grouping from some possible numbers in a sudoku cell as 1 and check the magical logic that implies. Optimally find a colour per word special 26000 or so word logic language. Then use this optimization ability to convert a llm into an interdependent numerical knowledge graph. Train it to work a chain of thought from a neuro symbolic dataset. Now you have better than o1 on a raspberry pi 5.
I know people who show high level of understanding the concepts of high stakes games between skilled opponents, but struggle to beat microstake games because they aren't able adjust to the fact their opponents don't play the same way. IMO the AI training against himself alone would suffer the same drawbacks initially. Understanding stupidity is an area I can imagine this type of AI struggling with. Even though the weaker players are highly predictable, their behaviour would have to be force fed to the learning process.
This AI tries to approximate a Nash equilibrium for poker by minimizing exploitability of it's strategy by iteratively attempting to find the best response to its own strategy, a stupid player shouldn't do better than experienced players against it
This is where science and real-life do not go well together. The AI was actually down in chips after playing 10,000 hands. But, after adjusting with variance-reduction (AIVAT) it was up in chips. It doesn't change the fact that it lost money.
Mostly agree. That check-raise on the river has GOT to be a bad play period, right? Your hand has WAY too much value to be raising as a bluff, and not nearly enough value to check-raise in the way it did. Literally the ONLY hand that the live player has that plays that way and might possibly call the raise is exactly QT (and sure.. QJ). I actually see this as a bad play by the robot in terms of an overall strategy. If your overall strategy has you check-raising a lot of top pair/mid kicker hands on the river, you're going to be losing a LOT of money.
@@MrAndersonmm I'm torn. It's generally a bad play to be check-raising top pair on the river, but this board isn't too scary. P6 checked the turn and therefore doesn't have a set. This river is as blank as it gets. The only hand I'd be worried about from AI pov is KQ, but even that I'd expect to see betting the turn for value. It's a really thin value check-raise that I don't think I would find, but that doesn''t mean it's losing.
Well , that was a bad poker hand analysis :) The double check is super standard with a top pair against the whole range of the raise. absolutely nothing special. And obviously the computer doesnt know the raising hand , could be AA KK AQ KQ .... and the hand would be similar
I reckon the human should have just bet turn there, in position theres no reason to give away free cards and you can get a handle on the strength of the opponent, the range of hands that would bet turn is pretty big. By checking back turn and small betting the river it looks like thin value. Not sure if checking turn would have been the right GTO play.
There is such a thing as an "optimal strategy" that includes the probability of slow playing in a given situation and while poker is a bit big of a game to solve this way, it is something that can be solved for numerically. There's also something called a "maximal strategy" that involves trying to model how your opponent differs from an optimal player and takes advantage of that. I don't find this so impressive as there is no sign that this is doing opponent modeling, this is just an approximation of optimal play - something that you can solve for with a large enough matrix solver.
Arrogant is the man who did an intro to AI class 2 years ago and thinks they can just "code this out real quick". I cant begin to comprehend this paper
So let me get this straight? A.I. have difficulty analyzing data and coming up with solutions when the information isn't presented whole(weakness),so you guys figured "Hey let's get rid of the only weakness A.I. so they can make Skynet a reality" People of the future, you have these guys to thank.
Is that 20 hours in wall-clock time or 20 hours in overall training time (i.e. 4 cores in parallel for 5 hours wall clock time)? If the latter, very impressive.
Wait, how did the computer bot know that it had a better kicker or that the other player didn't just have a two pair, or a three of a kind for that matter?
Queen with Jack kicker is going to be the best hand a huge percentage of the time so that's why it plays like that. In the long run the amount of times your opponent has a better hand than you will be so few that you'll make money overall
Understanding how the bot identifies a bluff or decides to fold would be key to defeating it. If a human player can simulate a bluff while holding a strong hand a winning strategy may begin to develop for a short time, and switching from that approach to conventional play at random intervals. At best it may result in a rock-paper-scissors distribution of wins, losses, and split pots. Also, how do these AI poker bots perform against themselves, and what strange gambling behaviors manifest?
First of all... at 2:47... that's NOT a slow play, a slow play is when you, etc have AA and someone is going all in, and you wait 5 minutes before you call the all-in. THAT'S what's defined as a slow play in poker. not checking all the way to the river and then do a reraise.
Not sure why the human didn't fold on the flop. It is a 2X pot raise, the pot odds are pretty bad. There are tons of combo that could have got there at the end. 3s and 2s because there wasn't a bet on the turn would have called the flop. QX combos. The bot can have a wild range playing in the BB. Honestly, the bot got lucky here, that gotta be the bottom range to call a 2x pot raise. If he was calling with Q10, he could easily be calling with QK, QA.
This works in online Poker. But (as someone who doesn't really play myself) isn't real poker much a game of psychology. Like reading your opponent and such? The AI would exceed in that, but it would feel like an unfair advantage as the AI doesn't have emotions that could "leak through"
I didn't watch it through but what I think will happen is machine's capacity to remember numbers at once is far greater than human's by design so, remembering all the passed cards plus never forgetting to compare a thing leads to beating even the most skilled human player, it's just math.
This isn't blackjack. There are no "passed cards". They play with 1 deck and reshuffle after every hand. Folded hands are put in the muck face down so you can't see them. The only cards you ever see until the hand is over are the 5 dealt face up and the two you have face down.
2:46 I just wanna understand the decision here:: there was time to modify the video (and re-encode) to correct a mistake of the voice over, but not re-record that part of the voice over? Was this Károly-bot's attempt to show weakness to fool us?
Correcting the audio would probably sound weird in context, because the segment will not be exactly as long as the original one was. Adding some corrective text is easier.
To anyone who is worried about it the simple answer is don't play against anyone or "thing" that can beat you. Poker pros don't make money playing each other they profit from weaker players. So if you are at a table where you do not have an edge you can leave that table or at least do not play against the player who is taking money off you. It's a no brainer- pick your spots and your opponents. It's all part of the game so bring it on poker bots! I make it a rule-don't play against anyone that has beaten me in the past.
I think the poker AI was not trying to trap the human. Top Pair J kicker is not that strong especially when someone open-raise pre-flop. Also, the QT has a backdoor flush draw out there on the flop.
All poker and casino sites have something in their terms and service about using certain types of programs........that being said if you get caught your not going to jail but you ip and account will be banned forever on the site. Also. They probably give this information out to other sites, so you could get blacklisted very quickly.
Online poker has been death for years buddy. It's been dominated by bots for atleast 5-10 years. You don't need this level of AI to destroy 99.9% of human players. However this only helps to push poker botting even further ahead of humanity
A bet is not a raise. I also have a hard time thinking about why anyone would check-raise QJ on the river. If the opponent has exactly QT he calls, if not you make worse hands fold and better hands call. I would even argue that QT could fold to a c/r on the rivet
I’m unsure how this AI would actually be really good, because humans should be able to predict the AI’s hands better given the AI will act more consistently in specific situations.
@@Vearru true, but that was just a view of the resulting logic as observed by humans for a particular scenario. It is extremely difficult to generalize and describe the behavioral logic of AIs from their "model" and not from the observational viewpoint.
@@HabibiGa1z because you'll be playing superhuman AI and guaranteed to lose. Unless you cheat and use the AI too, but that defeats the purpose of playing
Apparently if someone had hundreds of rounds you played poker they can use ai to play exactly like u.... And now there's websites selling big time poker players data. Scary
Winning the main event is a comparable achievement to winning the lottery; its almost entirely defined by luck. To give you an idea, the 2003 main event champion was a complete amateur called chris moneymaker. He just got super lucky thats all. The true world class players are the ones who regularly play the highest stakes at Pokerstars cash games (LlinusLlove, Otb_RedBaron, Trueteller, among others). They consistently win over large samples of hands such that the probability of them winning by pure chance is equal to 0%, thats why they are world class. The AI should have played against THEM, not against some degenerate who happened to get lucky at one big field tournament called the "main event".
Why not superhuman? Pluribus beat Linus Love, who is arguably the best 6-max player in the world. Beyond that, It looks like most of the human players that Pluribus faced were heads up specialists rather than 6-max specialists, but being the best of the best in the first format at least makes you a very decent player in the second.
@@synchronium24 most of the players (excluding LL obviously) really weren't the best of the best at anything, I analysed their stats from the hands made available and they played quite leaky strategies. The bot seemed to have a very good strategy from its stats. I don't think it played a statistically significant sample vs LL and as usual they are wilfully overstating the relevance of the achievements of the other players. Those other players would lose badly at a 5kNL table on pokerstars, I am confident of it. Beating them is by no means a superhuman achievement.
*Poker is al about bluffing* _Artificial Intelligence will use a algorithm based on the rest of the table their:_ - Seconds of thinking - Times raised when bluffen - How many percent raised - Times checked when insured - Seconds when somebody faults - Times faults - Chances of winning based on personal cards and cards on the table - ... *When you would play against “AI”, you need to play illogical, perhaps the machine uses advanced (illegal) software like:* - Infrared camera (to scan your temperature when you’re insured) - Voice scanner (to detect uncertainty while speaking) - Advanced camera (to detect body language) - ...
Im a midstakes online grinder, and i can tell you with certainty, The bot played The first Hand like an absolute fish in nl50 online would. He has no business donk betting qj on that River. Dont wna spend too much time writing this, but id say any1 whos any good at online cash would love to sit down with 5 of these bots...
I thought the computer power needed for nlhe was way too strong Limite hold’em has been mastered since way less variables are too count because bets are limited. If he is not lying on beating wsop winners and they played many hands that extremely impressive.
Weedbuiling Bs it is, what we’re seeing is the first interesting step into solving that as well. They continually set the stacks back to 100 big blinds because that’s all the AI is programmed to play at. It still can’t handle varying stack sizes etc. Still, it’s very impressive that it can give decent human players a run for their money 👍🏼
First get rich in online poker, then publish the paper
Which may have happened, we'll never know :)
This is the first thing i've thought about.
i think people who cheat do well at online poker ,if you're sat at a table where a few of the players are colluding then you're at a massive disadvantage , I think that's not so hard to do online
@@filippocastelli42
I could try an organic GAN by abducting Nocoldiz and Frederick Frankenstein, tied 'em up and make them work on the same video until it comes out perfect.
Don't know if that would be legal, though :)
@@ChristianIce probably not legal but morally justified
Ok.
Is it illegal to use an AI on online poker casino?
I ask for a friend.
I need it too, but not for a friend, for science! 😁
Well, I think there's no law that says the opposite
How could they catch you? Have a laptop running right next to you that isn't on the same network. RIP online poker.
This paper has been out for a month, I'm sure some smart people are quitely using it to their advantage already, likewise to when Snowden dropped the NSA info, till today hackers are using those vulnerabilities.
@@zukacs
They'd deserve every penny.
There, I said it :)
The AI clearly has the best poker face
gold
The AI also shows no body language, unlike puny humans.
You win
Funnily enough though, it doesn't. The AI only picks what it wants to do, not WHEN it wants to do it. So no really long wait to emphasize a bluff and make humans more antsy. No immediate raise to project strength (because everythin appears to be immediate.
A good play can use timings as well.
Yeah, I would argue that if you're not a human (or an alien with readable emotions) you're not actually playing Poker. The biggest reason for using money in Poker is to influence people emotions to make them more readable. Any other game with real stakes, statistics, and bluffing will play largely the same as Poker.
Always great to see papers with accompanying blog posts that make them accessible, while still explaining the broad ideas. Highly recommend anyone reads it if they haven't already.
How to get rich (funding for science): Step 1. Set up online poker. Step 2. Sell AI and let it battle.
So there's a lot of misinformation floating around about this A.I.
#1 the sample size was only 10,000 hands. Which to some of you may seem like a lot but to actual professional poker players they consider a sample size of at least 100,000 hands to see if you're a true winning player and to negate most variance.
#2 the bot had an extreme difficulty making computations with varying stack sizes and its performance worsened the more the stack sizes varied.
So, while impressive I would not consider it better than humans until I've seen more data
There seems to always be misinformation revolving around AI in general. Probably because people want to believe we are living in a science fiction movie were the evil machines take over the world
Agreed it’s preflop plays also seem kinda wack
I agree 10,000 hands is a joke. No way it accounts for variance. 100,000 isn't even enough. Even at most optimistic estimates, 300,000 is the minimum but most poker players will demand much more sample sizes.
Great for hit n run 6max shit
Pfft it's better. The higher the win rate, the less hands you need, and it was crushing the game. If it's not quite as good as stack sizes diverge, it can just quit and join another game.
Okay, I'm not a machine learning expert, but I don't understand where the "secret sauce" is in this paper. What, if anything, was stopping people from using this approach a long time ago? The paper mentions that the training algorithm isn't very different from ones used by previous poker AIs, and it didn't take a huge amount of money or specialized hardware (about $144 on commercial cloud services!) to train, or even to run (two high-end CPUs worth about $4,000). So why did nobody reach nearly this level of performance before?
I feel like I must be missing something important here. Usually, even if it's not clear *why* it works, I can at least get a sense of *what* was different. Here I just have no idea, but clearly they did something significantly better than anyone before.
You raise a good point! The success of the Pluribus project was not due to any major breakthroughs in machine learning algorithms or hardware, but rather to the careful integration of existing techniques and the development of a novel approach to game-playing.
One of the key innovations of Pluribus is its use of "nested subgame solving," a technique that allows the AI to reason about the game at multiple levels of abstraction and to make decisions based on the expected outcomes of subgames. This technique enables Pluribus to handle the complex strategic interactions between multiple players in a way that previous poker AIs could not.
Another important factor in the success of Pluribus is its ability to learn from its own gameplay and to adapt its strategies in real-time. This allows Pluribus to adjust its play style to exploit the weaknesses of its opponents and to avoid being exploited in turn. This kind of adaptive learning is a relatively recent development in machine learning, and it requires sophisticated algorithms and large amounts of data to work effectively.
Finally, the Pluribus project benefited from advances in cloud computing, which made it possible to train and run the AI using relatively inexpensive hardware and commercial cloud services. This reduced the cost and complexity of the project and made it more accessible to researchers and developers outside of large corporations or government agencies.
Overall, while the Pluribus project did not introduce any radical new techniques or technologies, it represents a significant step forward in the field of AI game-playing and demonstrates the power of combining existing techniques in innovative ways.
@@Swagtastic225 sounds like chatgpt
In "Example Hand 2", I don't think that huge check-raise on the River (4.7k raise in a 2k pot) with that hand was a good play. P6's call in response was also questionable, his hand was only beating bluffs at that point.
AI had good reason to think the human had nothing but a weaker pair-hand from the flop. Human player probably knew from many previous hands that the AI was unusually willing to make big bluffs in such situations (compared to most humans), and felt compelled to see if the annoying AI was trying to force him off a hand "yet again."
That was a very weird play by the AI. Most opponents would fold, I'm pretty sure I would. It looked way too much like it was trying to show weakness and then pretending to bluff.
Should've just bet the pot
>arguing with a superhuman AI
Good luck with that
The hand makes sense if you remind yourself these are tournament pros
In the advert, I like how you said WAN-D-B like it's a database, instead of W-AND-B for weights and biases lol
Online poker: exists
AI: I'll end this man whole career
Just writing a reply so RUclips ai would pick this comment up
@@StarOnCheek thanks
the most overused comment ever, let's put it on the top of the stack!
@@zlac that's how memes work
Lol that 'trap' with the QJ was a disaster, with that river raise you only get called by better hands. On top of that, humans trap their good hands all the time lol
I think that check back on the turn is a bad play, it basically puts the human on thin value when he bets the river.
@@lutherschultz4725 I disagree, as the human, you never get three streetz of value from worse hands there, so you either check back turn or check back river to control the pot
@@DaverendeDodo imo it's better to check back river than check back turn, because you let the opponent draw and limit your range on the turn when there is still action on the river. Unless if you include enough check backs on the turn with your tptk+ hands I think it's a bad play.
Exactly what I was thinking
Top poker pros rarely trap
The players have fixed stack size in this paper. If you consider real poker environment, player's stack can varies from 1bb to 300bb. Sure the most popular stack size is 100bb, but in real life there are can be different stack sizes, which can change strategy because you play with different potential pod odds
"What a time to be alive!" in the sense your uncle will have to quit online poker and will stop taking loans from Mom.
😂😂😂
Lol you just know they've been "testing" this in real rooms online for cash
So thats how they got the funding
Okay pluribus is pretty good although he played in a 6max format against players who weren't experts in 6max cash but in mtt's (tournament format) Which is pretty important.
Also I heard that after the end of every hand the stack sizes were reset to 100 big blinds. That means the bot doesn't know how to play short stacked or deep stacked poker only the standard 100bb. Also some of the hands played by pluribus were analysed in a gto (game theory optimal) solver and it basically showed that even the AI wasn't playing close to a perfect strategy.
I'd like to see a rematch but this time against the top 6 max players and without resetting the stacks. Only then you could say that AI can beat humans.
He played worse the more of differences in chips there were I’d like to see the bot go deep in a cash game
It played against Linus Loeliger (believed by many to be the best 6-max cash player at least at the time) and Nick Petrangelo (probably knows enough to beat some high stakes cash online even though he's a tourney player, arguably one of the best). I don't really know the other names in the list of players it played against, they're pros from slightly before the heavy GTO era e.g. Dong Kim I believe.
It wouldn't make any sense to play SnGs, tourneys or varying stack sizes to test the bot because then you would have extremely messy and meaningless results. Sng/Tourney is much higher variance and humans don't want to play for that long, and measuring its performance at 20bb, 40bb, 60bb, 80bb, 100bb, 120bb, etc would just be silly and subject to variance. Even in this low variance format, the results are noisy.
Pluribus plays preflop ranges very close to accurate preflop solves. If it deviates from what a solver says post-flop it probably does so in a way that doesn't blunder EV - I'd like to see an example of that if so.
Could you talk more about the variance reduction technique? Someone imported all the hands and found that the bot lost.
It was a genius play if you think about it. The back-check on the turn of P6 was actually really bad because the AI had to know that KQ or AQ is going to bet 3 streets. So either the AI gets a massive amount of value for having the higher kicker or it wins the pot anyway, because no weaker hand than QT is going to call the reraise on the river. I think its quite profitable in the long run.
If P6 would have c-betted the turn. The river would have been a check-check and P6 wouldnt have lost so much chips.
"This program was brought to you by Weights, and Biases" That reminds me of Sesame Street where they'd list numbers and letters of the alphabet as sponsors for the episode.
Thanks both of u ... always love these segments
I'd love to see the A.I's decision when Daniel Negreanu goes "You have K9"
I love AI playing games, it's amazing seeing it conquer game after game.
It'd be cool to see machine learning applied to video games like FTL or Slay the Spire
Same, I think it would be cool to play against a pro AI in any game so that maybe you could learn from it and get better from playing against it
Yeah or when proper bodies are developed then you can use them to learn something in any sport! Climbing, MMA, Football etc ...
Really liked your animation work on the poker table decision tree, well done Károly!
This is the death of poker ;) Nobody will play this once these AI's are around to dominate the game both online and live where people will cheat with earplugs ;)
Perhaps as there is so much money involved. Though I still find it unlikely as AI did not kill other games such as chess and go. We will just need to develop even better rules to ensure there is no cheating
@@justaguy7003 Yeah but online poker will be dead
AIs*
@@MrFram until convincing androids are made - Blade runner style.
MrFram Competitive online chess, go, etc aren’t dead. Why is this different? We have systems in place to detect people using bots to inform their decisions...
The hand that was shown can not be evaluated out of context. If a beginner showed me how he had won this hand and asked me for an evaluation, I would tell him that he overplayed his cards and was lucky to get called by pretty much the only worse hand that might pay him off. However, now that we know that the hand was played by a strong bot, that has established a table image of being a frequent bluffer, we can reevaluate the river check/raise as a brilliant merging of ranges. Knowing that it is capable of doing this, the humans can no longer consider a river check/raise just a very polarised bet that either means a total bluff or a super strong hand, like three of a kind, but now they also have to consider that it may be a medium strength hand, where they have to decide if their kicker is good enough.
This is scary. I am happy that I stopped playing online poker for money a long time ago. This is the official death blow to online poker, since it is now impossible to know if you are playing against a complete noob or a superhuman AI before you have lost so many hands that it is clear that it is not some random luck box.
I was completely with you until "This is the official death blow to online poker". It's going to be dangerous for pros playing each other, but most online players are still fish who neither are aware of Pluribus nor know how to use it.
this is the only sane analysis of the hand here
The highest levels of poker are usually face to face, where quick reads and expressive control play a big role. I wonder if there will be an AI that can make a heads up read and tell if a person is bluffing or has a strong hand. I think that could arguably be more powerful at a table, and scary!
I need help combining this with an image classifier and integrating into sunglasses with hidden camera.
rofl
You'd want a spy pen to bluetooth video to your phone, which would then upload it to a server that did the actual processing. An earbud could then tell you what to do. You'd still need a pokerface to bluff though :P
The line with QJ was a disaster. It almost felt like the bot knew that it dominates the opponent
I'm blown away by the quality here. I recently enjoyed a similar book, and it was top-notch. "Game Theory and the Pursuit of Algorithmic Fairness" by Jack Frostwell
would be so cool to pair this with a robot that could express facial emotions and read human player's faces
That excellent "slow-play" from the AI was actually a disastrous way to play poker....
I don't like your explaination for the A.I play... she can't know that her kicker is better as the player would, assuming he has a queen, raise pre-flop with Q-10 to AQ (so again assuming he has exactly one qeen it only beats 1 out of 3 hands he can have). The check post-flop from the A.I out of position is pretty standard, so is the c-bet from P6 and the call with top pair and possible straight draws. The A.I. checking that turn is again pretty standard, especially since the villain was the aggressor pre-flop and that you have top pair (so you want to induce a bet if you're opponent thinks he can bluff you of the hand. The check back from P6 makes sense since he has no more draws with the turn and the possibility that the A.I. is slow playing a better hand (AQ/QQ/KK/AA/99/77... although I assume the A.I would have 3-bet AQ/QQ/KK pre-flop), plus the fact that he wont get 3 streets of value from a worst hand. Now the check raise from the A.I. on the turn is really weird, especially considering the sizing of the raise, you would expect it to have either be a bluff or a monster, not a top pair 3rd kicker. It's very unlikely that the A.I.'s hand improved post-flop (the only hands that would be pocket deuces or threes. So if assuming the A.I. 3-bet AQ/AK/QQ/KK and maybe some smaller pairs pre-flop, the only hands in its very wide big-blind calling range that beat/tie you are AA/99/77/22/33/KQ/QJ/Q10/97(suited) and that's about it. As I said, I wouldn't expect such a raise from Q10 to KQ since you can easily be out-kickered by P6 so in my head the A.I has either one of the monsters I listed or one of the many more bluffs it could have check-called up to the river. For that reason (and unless we know the A.I. can slow playing hands like AQ/QQ/KK pre-flop out of position) the call makes some sense although you only beat bluffs.
It's really that huge raise from the A.I. that's weird and probably a bad play, has it will have most worst hands (exept some Q10) folding but most better hands (exept some QJ/KQ) calling (and monsters raising all-in)
As well as that I know that if I had a q 10 on the first hand and someone that was playing passively made a big raise I'd either re raise them if I was planning on buying in again or if not play it safe and cut my loses
You can't get any conclusion out of only one hand. According to the briefs "interviews" of the players who played against it, it was fond of bluffing, and was especially good on the river.
There is a game history that is not shown here (because it's a 5mn video talking about a paper and not a 1 hour video analyzing it's poker play), and it likely contributes to why this hand worked hat well.
That being said, I agree that this hand doesn't seem anything special to me.
Also, you can't rationalize how an AI plays using our human logic, it just doesn't think like that. And this is what is so interesting about it. In chess for example, since alphazero and leelazero games started to show off, some top-level way of playing and evaluating position has started to change a little bit, because the AI gave new ideas as to how to understand and evaluate positions.
This is the same here, if you stay stuck on the human way of understanding poker, you won't get the chance to improve with what the AI shows.
But once again, we can't extract anything meaningful out of only one hand, we agree on that.
I think the huge raise is a bad play only if you think that most people will fold to that. Maybe the AI has learned (and is able to recall) that 20% of the cases, people will call his raise and therefore the overall value made on the hand will be higher compared to making a smaller raise (that is called more often, but not often enough to make up for the difference in total money won). So that is why the river play seems crucial (at least according to this example hand).
@@Sanzarc The problem isn't the A.I. trying to get value but the hand it's trying to get value with, as it can be beat easily and that most of the time only worst hands will fold
@@brodieclamp5090 it wasn't the first hand, they start every hand with 10K chips to make things simpler for the AI
River check-raise seems a little bit thin for value but I suppose the button's range is somewhat capped after the turn check back. I just don't see a lot of worse hands that will call the raise (other than the exact QT button had). Maybe it's a bit of a merge play as you would fold out some KQ and maybe AQ while still getting value from a couple of worse Qx hands? I'd love to see the analysis because one hand where you have your opponent pipped doesn't show much.
totally agree, wanna see more hands
the problem with this scenario is that whatever test subjects they used are likely not playing at the level they would be if the stakes were high (bracelet and purse)
Please tell us who are the two wsop main event players the AI played against. Surprising to see that none of those players admitted anywhere that they actually played this game against AI
How do we know this isn't a trick by online Poker companies to try and get us to play their online Casinos? And then the bot makes us lose all our money????? 😱😱😱
Well, duh, you use the bot yourself.
You don't go to poker sites expecting to win money if you're not using an ai
Also, bots have infested poker sites for quite some time anyway, it's not exactly a new strategy.
@@ekkehard8 Well, yeah that's if the bot actually works and isn't a trick! 🤔
Cause we dont got that bot
If you play on any legitimate site I wouldn’t worry about it. Besides, any readily available ‘bot’ that can play for you undetected on a site these days would actually be losing money
Wow, that example hand was nuts. Mainly because any good poker player knows that you generally go aggressive there and raise. Not raising until the end is thought to be super risky. Generally, you just want to win your pot ASAP. Very interesting stuff!
FYI in some of the examples shown both flop and river play by human were bets not raises. However I am sure the bot knows this after all only humans make mistakes
The paper is behind a paywall. Has anyone found a free version online?
www.cs.cmu.edu/~noamb/papers/19-Science-Superhuman.pdf
@@aydingerek4726 Thank you!
Great find. Adding to the video description. Thank you!
Note a possible grouping from some possible numbers in a sudoku cell as 1 and check the magical logic that implies.
Optimally find a colour per word special 26000 or so word logic language.
Then use this optimization ability to convert a llm into an interdependent numerical knowledge graph.
Train it to work a chain of thought from a neuro symbolic dataset.
Now you have better than o1 on a raspberry pi 5.
The secret to beating an AI at chess is to never let them see your pieces
Most underrated comment
This is dangerous, imagine this can be uploaded in a tiny computer in an eyeglasses and the wearer enters a casino.
I would be curious to know how it performed against casual players.
I know people who show high level of understanding the concepts of high stakes games between skilled opponents, but struggle to beat microstake games because they aren't able adjust to the fact their opponents don't play the same way. IMO the AI training against himself alone would suffer the same drawbacks initially. Understanding stupidity is an area I can imagine this type of AI struggling with. Even though the weaker players are highly predictable, their behaviour would have to be force fed to the learning process.
A usual poker table has 8 positions available, Let's try that again a d include a beginner as well as a professional.
This AI tries to approximate a Nash equilibrium for poker by minimizing exploitability of it's strategy by iteratively attempting to find the best response to its own strategy, a stupid player shouldn't do better than experienced players against it
This is where science and real-life do not go well together.
The AI was actually down in chips after playing 10,000 hands. But, after adjusting with variance-reduction (AIVAT) it was up in chips.
It doesn't change the fact that it lost money.
"But, after adjusting with variance-reduction (AIVAT) it was up in chips. "
I'm not familiar with this term. Is it different from all-in-adjusted EV?
The QJ vs QT hand really smells like Mike Postle kind of "great play", especially with the super-sketchy explanation you gave. Not the best example.
Mostly agree. That check-raise on the river has GOT to be a bad play period, right? Your hand has WAY too much value to be raising as a bluff, and not nearly enough value to check-raise in the way it did. Literally the ONLY hand that the live player has that plays that way and might possibly call the raise is exactly QT (and sure.. QJ). I actually see this as a bad play by the robot in terms of an overall strategy. If your overall strategy has you check-raising a lot of top pair/mid kicker hands on the river, you're going to be losing a LOT of money.
@@MrAndersonmm I'm torn. It's generally a bad play to be check-raising top pair on the river, but this board isn't too scary. P6 checked the turn and therefore doesn't have a set. This river is as blank as it gets. The only hand I'd be worried about from AI pov is KQ, but even that I'd expect to see betting the turn for value.
It's a really thin value check-raise that I don't think I would find, but that doesn''t mean it's losing.
Well , that was a bad poker hand analysis :) The double check is super standard with a top pair against the whole range of the raise. absolutely nothing special. And obviously the computer doesnt know the raising hand , could be AA KK AQ KQ .... and the hand would be similar
Thanks for pointing that out so I don't have to :)
I reckon the human should have just bet turn there, in position theres no reason to give away free cards and you can get a handle on the strength of the opponent, the range of hands that would bet turn is pretty big. By checking back turn and small betting the river it looks like thin value. Not sure if checking turn would have been the right GTO play.
Now I don't have to write this. Thx :)
There is such a thing as an "optimal strategy" that includes the probability of slow playing in a given situation and while poker is a bit big of a game to solve this way, it is something that can be solved for numerically. There's also something called a "maximal strategy" that involves trying to model how your opponent differs from an optimal player and takes advantage of that.
I don't find this so impressive as there is no sign that this is doing opponent modeling, this is just an approximation of optimal play - something that you can solve for with a large enough matrix solver.
Gonna pop open tensorflow real quick and code this out.
Arrogant is the man who did an intro to AI class 2 years ago and thinks they can just "code this out real quick". I cant begin to comprehend this paper
" My Poker AI is more talented than yours! " :-)
What a time to be alive
I could finally win at Poker Night at the Inventory if I had this AI…
I always wanted Strong Bad’s “Dangeresque Too” glasses…
Poker Night is easy. Just play TAG - play less hands, and when you get a good hand, bet BIG. They'll often call you down with jack shit offsuit.
Yooo the A.I knows how to bluff and slow roll??? That's crazy
Bluffing is crucial to be a winning poker player :)
So let me get this straight? A.I. have difficulty analyzing data and coming up with solutions when the information isn't presented whole(weakness),so you guys figured "Hey let's get rid of the only weakness A.I. so they can make Skynet a reality" People of the future, you have these guys to thank.
Is that 20 hours in wall-clock time or 20 hours in overall training time (i.e. 4 cores in parallel for 5 hours wall clock time)? If the latter, very impressive.
Doesnt mater, that still wouldnt tell anything about how powerful the cores (or thousands or cores of GPUs more likely) are
Wait, how did the computer bot know that it had a better kicker or that the other player didn't just have a two pair, or a three of a kind for that matter?
Queen with Jack kicker is going to be the best hand a huge percentage of the time so that's why it plays like that. In the long run the amount of times your opponent has a better hand than you will be so few that you'll make money overall
Classification and Regression trees were developed in the field of Statistics
We have chess AI tournaments.
Next is poker AI tournaments.
Understanding how the bot identifies a bluff or decides to fold would be key to defeating it. If a human player can simulate a bluff while holding a strong hand a winning strategy may begin to develop for a short time, and switching from that approach to conventional play at random intervals. At best it may result in a rock-paper-scissors distribution of wins, losses, and split pots.
Also, how do these AI poker bots perform against themselves, and what strange gambling behaviors manifest?
did you played rock papper scissors ai..!??
its realy hard to win
Does the Bot know how many cards are in the deck and which cards are out?
An AI could never beat an unpredictable poker player. Period.
you mean a noob? lol. yah, but noobs grow up and start to lose fast
Try to play rock papper scissor ai and change your mind.
That QJ hand was horribly played by AI. Positive results don’t mean positive EV play
First of all... at 2:47... that's NOT a slow play, a slow play is when you, etc have AA and someone is going all in, and you wait 5 minutes before you call the all-in. THAT'S what's defined as a slow play in poker. not checking all the way to the river and then do a reraise.
Not sure why the human didn't fold on the flop. It is a 2X pot raise, the pot odds are pretty bad. There are tons of combo that could have got there at the end. 3s and 2s because there wasn't a bet on the turn would have called the flop. QX combos. The bot can have a wild range playing in the BB. Honestly, the bot got lucky here, that gotta be the bottom range to call a 2x pot raise. If he was calling with Q10, he could easily be calling with QK, QA.
This works in online Poker. But (as someone who doesn't really play myself) isn't real poker much a game of psychology. Like reading your opponent and such? The AI would exceed in that, but it would feel like an unfair advantage as the AI doesn't have emotions that could "leak through"
Wonder if an AI also will br able to play a very simple game such as Yatsi in a way that it wins most of the times ?
Being a main event champion doesn’t mean you’re a top tier player. I hate that they act like that’s a big deal
I didn't watch it through but what I think will happen is machine's capacity to remember numbers at once is far greater than human's by design so, remembering all the passed cards plus never forgetting to compare a thing leads to beating even the most skilled human player, it's just math.
This isn't blackjack. There are no "passed cards". They play with 1 deck and reshuffle after every hand. Folded hands are put in the muck face down so you can't see them. The only cards you ever see until the hand is over are the 5 dealt face up and the two you have face down.
If u see AI does a thin value raise on the river like that next time u are in that spot u can trap him with weak river cbet with a stronger hand
Saying raise instead of bet tilted me for the whole video
?? Does it matter? Same shit bro.
I was just wondering...is it possible to crate AI that would optimized codes? or maybe that s already available?
i always wonder how you make something like this very impressive
Poker is cool but what about Mafia? With text generating ai algorithm. It would be a real challenge to achieve.
2:46 I just wanna understand the decision here:: there was time to modify the video (and re-encode) to correct a mistake of the voice over, but not re-record that part of the voice over?
Was this Károly-bot's attempt to show weakness to fool us?
Correcting the audio would probably sound weird in context, because the segment will not be exactly as long as the original one was. Adding some corrective text is easier.
Soon there will be no games left for us humans
You can gain some comfort from the fact that games are designed to be challenging for humans.
Just as you can't walk because there are cars...
The mathematical aspect of poker that an AI can excel in is only half of the game.
I still think professional players would smoke this bot.
To anyone who is worried about it the simple answer is don't play against anyone or "thing" that can beat you. Poker pros don't make money playing each other they profit from weaker players. So if you are at a table where you do not have an edge you can leave that table or at least do not play against the player who is taking money off you. It's a no brainer- pick your spots and your opponents. It's all part of the game so bring it on poker bots! I make it a rule-don't play against anyone that has beaten me in the past.
thanks
I think the poker AI was not trying to trap the human. Top Pair J kicker is not that strong especially when someone open-raise pre-flop. Also, the QT has a backdoor flush draw out there on the flop.
Playing online would not be great because Pluribus takes 40 CPU cores to run. You would be better off mining bitcoin with that computing power.
Less advanced versions are used in low stake tables and are profitable
All poker and casino sites have something in their terms and service about using certain types of programs........that being said if you get caught your not going to jail but you ip and account will be banned forever on the site. Also. They probably give this information out to other sites, so you could get blacklisted very quickly.
Now the Terminator will crush Jon Connor in a game of poker too.
Does this mean that online poker is dead?
We are late. :D
Online poker has been death for years buddy. It's been dominated by bots for atleast 5-10 years. You don't need this level of AI to destroy 99.9% of human players. However this only helps to push poker botting even further ahead of humanity
@@nonfiction876 at least*
@@JorgetePanete Thanks, buddy. Your vigilance in pointing out inconsequential typos on every post is much appreciated.
No, just that you have to make the better AI
A bet is not a raise. I also have a hard time thinking about why anyone would check-raise QJ on the river. If the opponent has exactly QT he calls, if not you make worse hands fold and better hands call. I would even argue that QT could fold to a c/r on the rivet
I’m unsure how this AI would actually be really good, because humans should be able to predict the AI’s hands better given the AI will act more consistently in specific situations.
The AI will take that into account and either stop being consistent or, whats worse, trap humans into that delusional consistency
Alex Tikh okay, based off the decision tree that was shown that didn’t seem to be the case, but hopefully it would be able to be do that.
@@Vearru true, but that was just a view of the resulting logic as observed by humans for a particular scenario. It is extremely difficult to generalize and describe the behavioral logic of AIs from their "model" and not from the observational viewpoint.
Feel the pain - greets from FX traders
So who already set this up on an online poker site?
And now online poker is no longer a thing
what u mean
@@HabibiGa1z because you'll be playing superhuman AI and guaranteed to lose. Unless you cheat and use the AI too, but that defeats the purpose of playing
@@Jone952 well u said ur talking a few years ahead and not now
@@HabibiGa1z If this AI is currently available for download I'd stop playing immediately
@@Jone952 well let me know if u find out if it is
I'm pretty sure people made some pokerbots, even "among us" bots despite the lack of money to win.
2:49 "excellent play for the AI"
it's a very normal play for anyone that had played poker for more than 3 hours ..
"excellent play *for the AI*"
he says at the beginning that Ai has a difficult time with Poker
I cant find this hand in the 5H1AI logs.
How to start it? My notebook is downloaded by file ...
Apparently if someone had hundreds of rounds you played poker they can use ai to play exactly like u.... And now there's websites selling big time poker players data. Scary
Winning the main event is a comparable achievement to winning the lottery; its almost entirely defined by luck. To give you an idea, the 2003 main event champion was a complete amateur called chris moneymaker. He just got super lucky thats all.
The true world class players are the ones who regularly play the highest stakes at Pokerstars cash games (LlinusLlove, Otb_RedBaron, Trueteller, among others). They consistently win over large samples of hands such that the probability of them winning by pure chance is equal to 0%, thats why they are world class. The AI should have played against THEM, not against some degenerate who happened to get lucky at one big field tournament called the "main event".
Mike Postle can crush this AI easily
I mean he is the God of poker after all
It's not superhuman. It's still a very impressive AI but the claims made by the researchers are completely disproportionate.
Why not superhuman? Pluribus beat Linus Love, who is arguably the best 6-max player in the world. Beyond that, It looks like most of the human players that Pluribus faced were heads up specialists rather than 6-max specialists, but being the best of the best in the first format at least makes you a very decent player in the second.
@@synchronium24 most of the players (excluding LL obviously) really weren't the best of the best at anything, I analysed their stats from the hands made available and they played quite leaky strategies. The bot seemed to have a very good strategy from its stats. I don't think it played a statistically significant sample vs LL and as usual they are wilfully overstating the relevance of the achievements of the other players. Those other players would lose badly at a 5kNL table on pokerstars, I am confident of it. Beating them is by no means a superhuman achievement.
Show face of person in camera to ai that will change game to another level
How does it use timing when bluffing if it was only playing against itself in sped up matches?
*Poker is al about bluffing*
_Artificial Intelligence will use a algorithm based on the rest of the table their:_
- Seconds of thinking
- Times raised when bluffen
- How many percent raised
- Times checked when insured
- Seconds when somebody faults
- Times faults
- Chances of winning based on personal cards and cards on the table
- ...
*When you would play against “AI”, you need to play illogical, perhaps the machine uses advanced (illegal) software like:*
- Infrared camera (to scan your temperature when you’re insured)
- Voice scanner (to detect uncertainty while speaking)
- Advanced camera (to detect body language)
- ...
Im a midstakes online grinder, and i can tell you with certainty, The bot played The first Hand like an absolute fish in nl50 online would.
He has no business donk betting qj on that River. Dont wna spend too much time writing this, but id say any1 whos any good at online cash would love to sit down with 5 of these bots...
I thought the computer power needed for nlhe was way too strong
Limite hold’em has been mastered since way less variables are too count because bets are limited. If he is not lying on beating wsop winners and they played many hands that extremely impressive.
Weedbuiling Bs it is, what we’re seeing is the first interesting step into solving that as well. They continually set the stacks back to 100 big blinds because that’s all the AI is programmed to play at. It still can’t handle varying stack sizes etc.
Still, it’s very impressive that it can give decent human players a run for their money 👍🏼
Why gives an example where AI has stronger hand? What if the other has a set? Also good player will make a good fold with just top pair Q