Sign Up For Card Player's Newsletter And Free Bi-Monthly Online Magazine

Poker Bot Expert Steve Blay Goes All-In On Omaha Poker Training... And Wins!

Omaha Poker Training Is The Latest Game Development Tool From Advanced Poker Training


Steve Blay has been programming bots for poker training for more than 10 years. When he developed Advanced Poker Training, teaching bots to play no-limit hold’em realistically was not an easy chore. However, with just two hole cards, it was a manageable problem.

Blay apparently had not faced enough challenges in his poker programming career, so in 2018 he decided to take on the geometrically more complicated game of pot-limit Omaha (PLO). Nearly two years, multiple consults with leading PLO experts, and dozens of revisions later, Blay brought Omaha Poker Training to fruition.

Card Player Magazine sat down with Blay and discussed everything from machine learning, to whether poker bots play online for real money, to how a no-limit hold’em player can make the leap to PLO – where the pots are huge and the action is wild!

Card Player: Advanced Poker Training (APT) must take up a lot of your time. How did the idea of developing a training simulation for PLO first develop?

Blay: Well, first I determined I wasn’t busy enough doing private coaching, running, raising a couple of kids, etc. But seriously, I’ve gotten about a thousand emails asking for it over the last 10 years. I’m probably not even exaggerating very much there. It’s at least 500.

Card Player: What struck you as the greatest challenges you’d face in creating bots that play like real PLO players?

Blay: The same challenges I faced in creating APT – namely, getting a computer program to “see” the game like a human sees it, and make human-like mistakes. In other words, not just trying to use brute-force math, but adding some human psychology.

In a sense, it’s easier to write GTO-style bots. Anyone who has followed the chess world knows that the best chess programs have been able to beat every human on earth for over 20 years now. With so much computing power, that’s trivial today. Playing against an opponent like that isn’t of much value to the typical player. You’ll just get destroyed, and you won’t learn much.

Card Player: I actually play chess, and I hate playing against the computer – it always wins.

Blay: Exactly. To create a chess program that plays like a human, one that makes truly human-like misjudgments that can serve as a good sparring partner for a human – now that’s a real challenge. We’ve only seen real progress on that problem over the last few years with the advances in machine learning. Now, you can feed a computer millions of chess games played by amateurs and have it learn to play like an amateur.

But back to (OPT). What I’ve tried to do is offer the best of both worlds. For the GTO crowd, the bots really do play at a high level. You can look at the exact range of starting hands they are holding – all 270,725 of them – if you really want. But for beginners and intermediate players, you can dial down the difficulty substantially. This is where the real magic happens, because you can train against lifelike bots that are going to play much like the typical players you meet in a cardroom.

Card Player: That’s my favorite thing about Advanced Poker Training.

Blay: Thanks. And no other site offers that. There are other places to play no-limit hold’em against bots, but they lack the realism. And, of course, APT also has the “Beat the Pro” Challenges, the Combat Trainers, and everything else that has made it so successful.

Card Player: You were able to talk to some premier PLO players about their approaches to the game while programming OPT. Who did you speak with, and what kind of insights did they have to offer?

Blay: I’ve worked with Scott Clements on and off for years and have come to the conclusion that he’s one of the best minds in poker, and clearly one of the best in the world at PLO. So it was natural to enlist some help from him. But, he’s not really a GTO guy. He’s more instinct and play-the-player. I really wanted to build some GTO into these bots too. For that part, I consulted with John Beauprez. He’s not a household name but he has a whole team that has analyzed PLO since solvers were first developed, so they were of great help with the GTO aspect.

Card Player: What would you advise a no-limit hold’em player in terms of what they need to be a successful PLO player?

Blay: First, get used to big pots and lots of action. Don’t fear it, learn to love it. Second, concepts like position, blockers, and redraws are critical in PLO. Get used to counting your outs quickly – a great draw in PLO can have 20 or more. Finally (and seriously), join, unless you like to lose thousands of dollars at the cardroom learning to play the game well.

Card Player: In no-limit hold’em there are teams of people tackling the game and devising GTO bots. That doesn’t seem to exist for PLO – which makes your site unique. Why do you think there aren’t any good computer-based PLO games?

Blay: It’s just really hard. The game of PLO is exponentially more difficult than no-limit hold’em. Consider this: suppose you are heads-up on the flop. How many different combinations are there of your cards, your opponent’s cards, and the flop cards? For PLO the answer is roughly 700 TRILLION. And that’s heads-up! There isn’t anything like OPT because it is incredibly difficult to create. I worked daily on this for nearly two full years, and that was with previous experience building the APT no-limit hold’em bots.

A secondary problem is that the bots have to be able to come to a decision in a reasonable amount of time. Sure, anyone can make PLO bots if you give them an hour per decision to think through all the combinations. I set as one of the requirements that the bots have to act in two seconds or less. Anything longer, and it cuts down on how fast you can train.

Card Player: So pure number crunching power and speed are what make a good PLO bot?

Blay: Well, like I said before, what makes a good PLO bot for training purposes is realism. In 2020, we have an incredible amount of computing power available literally at our fingertips. Teaching bots to crunch numbers is one thing. But if the bots only think that way, their play will be unimaginative and mechanical. Therefore, from a practical sense, the best bot makes human-like misjudgments from time to time. That’s what I spent the most time on.

Card Player: What were you most surprised by when programming the OPT bots?

Blay: My favorite thing is when I think a bot is playing a hand terribly, only to find out on a later street that he had a devious plan all along. I didn’t teach these bots to think in the same way I did on APT. On APT there are very specific “plays” programmed in – the continuation-bet, the check-raise, the blocking bet. That doesn’t mean the APT bots are inferior, just different. On OPT, in a sense the bots came up with much of their strategy on their own. They taught themselves. On the hardest level they can be so frustrating, but in an awesome way. It brings a huge smile to my face when I get owned by them, because it’s a thing of beauty that they are better than their programmer.

Card Player: When a player is using your simulator, what kind of advice or guidance do they get from your site to inform their decisions during a hand?

Blay: Just about anything you could possibly want to know. For beginners, it will help you compute the pot size, see your odds of improving, pot odds, stack-to-pot ratio, and more. We also have a series of instructional videos and articles that can teach a complete novice how to play the game. I know there are so many good no-limit hold’em players who would love to play the big pot game of PLO when they walk into a cardroom and see a loose game, but don’t know where to start. This is their chance.

For experts, you can actually look at your entire hand range, and the ranges of your opponents, to perform detailed analyses.

Card Player: What are the key improvements you think PLO players will be able to make using your training tool?

Blay: The beauty of OPT is that every hand you play is saved forever into your own online database. You can get detailed reports about your play filtered by a huge number of criteria, including dates, hand types, and position. After a while, you’ll start to see trends in your reports. You can focus on your trouble areas by configuring the software to deal you specific kinds of hands. You can freeze the button and only play out of the big blind, for example. You can deal yourself only playable or raising hands. You can dial up the aggressiveness of your opponents if those loose-aggressive types are giving you trouble. It’s focused practice that really cuts down your training time.

Card Player: So the question I’m sure everyone is wondering is, have you let these bots play online for real money?

Blay: No, besides it being unethical, I would have no idea how to do that. There’s a lot of hype about bots playing for real money online, but I personally don’t know anybody successfully doing it. It takes very specialized software to read and process all the information on the screen and “robotically” click the correct buttons. And of course, your bots have to be good enough that they don’t actually lose your money!

Finally, the online poker sites are always working on countermeasures, and if you get caught you get locked out and there goes all your investment. So, while I wouldn’t say there aren’t any bots playing online, it isn’t rampant, it requires constant oversight, and frankly there are easier ways to make money.

Card Player: What have you yourself learned as a result of this process?

Blay: Geez, what haven’t I learned? The beauty of this whole thing is that I was not an expert PLO player when I started this project. After playing hundreds of thousands of hands against these bots over the past couple of years, I can play comfortably in about any PLO game. I think everyone ought to be able to play PLO, because sometimes you go to a cardroom and all the loose, crazy players are at the PLO table, and you want to be able to jump right in. Frankly, the players are generally much worse in PLO and the pots are bigger.

Card Player: What is your favorite thing about Omaha Poker Training?

Blay: Hands down, the retry feature. Any hand you play, you can try it over again as many times as you want, to test out different strategies. I also really like this feature we call the “grid”, which lets you take an overall view of all the hands in your range. Especially in bluffing situations, it lets you determine what hands you can represent believably. If you have trouble bluffing successfully, this could be your best friend.

Card Player: So what about Advanced Poker Training, does this replace it? Is hold’em dying?

Blay: Not at all. APT will always be our flagship product and continues to be the world’s no. 1 poker training site. We’ve been adding content to that site for 10 years, and there’s nothing else like it out there. We continue to have a team, including me, working on it every day.

As far as hold’em dying, I recently went to the WSOP Circuit event in Tampa. The first event had a $500,000 guarantee. So many people showed up that they were running out of tournament chips. The prize pool ended up being $1.25 million. So if hold’em is dying, nobody told Tampa.

Card Player: What’s next? Should we expect a Badugi simulator coming soon from APT?

Blay: Oh man, that would be fun. Now you’ve got me thinking. ♠