Hello Reddit! We are Jason Les and Dong Kim, part of a 4-person team of top professional poker players battling Libratus, an AI developed by PhD student Noam Brown and Professor Tuomas Sandholm at Carnegie Mellon University. We are among the best in the world at the form of poker we're playing the bot in: Head's Up No-Limit Texas Hold'em. Together, we will play 120,000 hands of poker against the bot at the Rivers Casino, and it is all being streamed live on Twitch.

Noam and Dr. Sandholm are happy to answer some questions too, but they can't reveal all the details of the bot until after the competition is over.

You can find out more about the competition and our backgrounds here: https://www.riverscasino.com/pittsburgh/BrainsVsAI/

Or you can check out this intro video: https://www.youtube.com/watch?v=JtyA2aUj4WI

Here's a recent news article about the competition: http://gizmodo.com/why-it-matters-that-human-poker-pros-are-getting-trounc-1791565551

Links to the Twitch streams:

Jason Les: https://www.twitch.tv/libratus_vs_jasonles

Dong Kim: https://www.twitch.tv/libratus_vs_dongkim

Jimmy Chou: https://www.twitch.tv/libratus_vs_jimmychou

Daniel McAulay: https://www.twitch.tv/libratus_vs_danielmcaulay

Proof: http://www.cs.cmu.edu/~noamb/brains_vs_ai.jpeg https://twitter.com/heyitscheet/status/825021107895992322 https://twitter.com/dongerkim/status/825021768645672961

EDIT: Alright guys, we're done for the night. Thanks for all the questions! We'll be playing for three more days though, so check out the Twitch tomorrow!

EDIT: We're back for a bit tonight to answer more questions!

EDIT: Calling it a night. Thanks for the questions everyone!

Comments: 1074 • Responses: 91  • Date: 

krnhydra1770 karma

This is a question for Dong. Are the other three guys going to chip in to pay for a massage for your back, which I assume is super sore from carrying the team for 20 days?

brains_vs_ai992 karma

Dong: I hope so.

qCrabs434 karma

Won't this destroy online poker?

brains_vs_ai751 karma

Its just a matter of time. It was a good run

JoeyMafia364 karma

1) I've seen the bot turn very strong hands into bluffs in spots that I wouldn't really think it would as well just jam with nothing and no real relevant blockers. Do you think that the bot has been trying to exploit the humans in spots where it believes that the human range is capped or do you just think it's playing GTO and remaining balanced?

2) I've seen Jason increase pre flop raise sizing as the challenge has progressed. What's the reason behind this + do you think it's been helping?

3) How do you think the bot would fair against the best in the world (OTB, Sauce, Doug)? Not that you guys aren't awesome.

4) Do you ever tank on every action to let the computer feel some of it's own medicine?

Love the streams. Can't believe this shit doesn't get more viewers.

brains_vs_ai358 karma

1) Dong: The bot doesn't give a shit about its range being capped. Second, I think it just picks a frequency and categorizes its checks and bets into different sizes. Jason: I think it's very rarely not taking blockers into account. It seems to always take blockers into account in every situation.

2) Jason: We've all been increasing our preflop sizes. It's our attempt to take advantage of any perceived weakness in Libratus. Dong: It's hard to tell if it's helping.

3) Dong: I think it would be very similar results regardless of which pro was playing.

4) Dong: Sometimes I have the nuts and bot jams on me and I go get a coffee so I can come back to a good situation. For the most part though, I come back to a really shitty situation.

Boocks337 karma

This is a question for Dong and Jason. In terms of how the computer plays would you say it's like playing a very strong human player or is it playing in a different way to how a human would play?

brains_vs_ai539 karma

Jason: We're seeing the bot play like a strong human player, but also putting way more pressure on us than any human can correctly.

HeywoodUCuddlemee312 karma

Have you considered bringing on Will Kassouf to try the speech play angle?

brains_vs_ai268 karma

Jason: We're willing to try anything at this point

raptor08191 karma

Noam, after the challenge, it would be great if you could do an AMA to answer some questions we have during the challenge about Libratus' strategies and learning and adjustment capabilities. Would you and/or Prof. Sandholm be willing to do that?

brains_vs_ai203 karma

Noam: I'd definitely be interested in that. Maybe a Science AMA.

ChemEBrew13 karma

Noam: can you say if the bot is using a supervised ANN or is it continuous learning? Or no machine learning at all?

brains_vs_ai32 karma

Noam: The basis for the bot is reinforcement learning using a special variant of Counterfactual Regret Minimization. Prior to this competition, it had only played poker against itself. It did not learn its strategy from human hand histories.

ChemEBrew16 karma

Now as I understand in CRM, the AI plays a hand against itself, and it makes a decision during its play. After the result, it reevaluates the acted on decision. Is it possible in one variation of this algorithm to use a Monte Carlo approach to create several hypothetical decisions in a play, choosing one central decision, and then evaluating the distributed hypothetical plays to learn faster? I hope I'm wording this correctly.

brains_vs_ai27 karma

That's sort of what we're doing actually. We use a form of Monte Carlo CFR distributed over about 200 nodes. We also incorporate a sampled form of Regret-Based Pruning which speeds up the computation quite a bit.

Tribunal_141 karma

This is to Dong.

What would your thoughts be, if they after the challenge revealed that you didn't play against the bot but played against Doug Polk?

brains_vs_ai386 karma

Dong: I would be very surprised, because Doug Polk was shooting a video behind us while we were playing. That's some impressive multitasking.

DaveShoelace141 karma

How do you think the outcome of this match will affect the future of poker?

brains_vs_ai393 karma

Jason: The fact is the AI exists, and it's extremely tough whether we win or lose. Even if we were to just barely win, it would have been an AI that could beat essentially everyone. This is going to be a problem for internet poker as time goes on.

Marshy92109 karma

How excited are you guys to go crush some regular human opponents after trying to grind it out against Skynet? More seriously: how has your strategy improved from facing an opponent like Libratus? Are you gonna start overbetting more frequently you think?

brains_vs_ai149 karma

Jason: Once you face Libratus, there's nothing worse any human could ever do to you. Every human is going to seem like a walk in the park.

Jason + Dong: We are definitely going to start overbetting more frequently. It takes a lot of studying to figure out the right way to do it though. The moment you're somewhat imbalanced there (bluffing too much, or bluffing too little) then you're making a huge mistake.

WouterDeLeur106 karma

Why are you all opening to 6x from the button?

brains_vs_ai276 karma

Dong: Desperate times, desperate measures.

lapp3r3035 karma

So isn't that just a way of trying to get on the good side of variance? If it's beating you at a reasonable rate aren't you just exacerbating the situation?

brains_vs_ai74 karma

Jason: We have strategic reasons for doing it, not just trying to make the best out of the good side of variance. You're correct that it could make the situation worse but at this point it's really our best hope.

DaveShoelace86 karma

Do you guys think that if you had a year to review the hand histories, that you could beat this current version of the AI?? Or at least come close?

brains_vs_ai176 karma

Jason: I think if we had infinite time to study and play on any schedule we wanted, we could get closer. But I don't think we could beat it.

oskar66986 karma

What is the dumbest thing you've seen the AI do so far?

brains_vs_ai198 karma

Jason: I opened JJ it called. Flop 872 rainbow, I cbet about 2/3 pot and it went all in for 200bb with KTo. This isn't necessarily "dumb" but it was quite "WTF"

thedeliberative15 karma

"who" won the hand?

brains_vs_ai28 karma

Jason: When we go all-in, we just split the pot according to our equity our hands have vs each other. So, I won the majority of the pot.

brains_vs_ai58 karma

Folded Kings preflop to a 4bet or 5bet... It knew....

Tribunal_68 karma

This is to Jason and Dong.

  1. Do you tend to bluff more against the bot than against a human player?

  2. What would you say is the bots strongest game? Preflop, postflop, turn or river?

brains_vs_ai156 karma

1) Jason: No, not at all. You have to be very careful with your bluffs against Libratus because he's not subject to emotion like humans are. Dong: There are humans that fold 100% in some situations, or never make a bluff in a scary situation. But the bot doesn't care. Humans have an unwritten rule that they won't bluff in some spots, but the bot does it anyway.

2) Dong: Turn.

CaseJr65 karma

Do any of you feel you've learnt or are learning things & improving your own game by playing this AI bot? If you'd like to elaborate, feel free, but I understand the need to remain as mysterious as possible.

brains_vs_ai126 karma

Jason: Yeah definitely. Playing Libratus has really forced me to push my poker knowledge to the limit in every situation. Plus, the best way to get better at poker is to play people who are better than you.

jakeputz57 karma

Do you think it might be possible to tilt the AI by calling it a "bad reg" or with a well timed "BAZAM"?

brains_vs_ai38 karma

shout out to upswingpoker :D

arcangel09246 karma

How often does the bot donk lead?

brains_vs_ai55 karma

about 14%

adamchikas42 karma

Are these robots heads up only? Are there robots who can play tournaments?

brains_vs_ai65 karma

Noam: The bot is currently only heads up. A lot of the methods can be applied to games like 6-max, but we haven't really tried that yet.

SernyRanders39 karma

What's the bots xr frequency f/t/r?

brains_vs_ai49 karma

Dong: Around 9, 8, and 12.

abusepotential38 karma

Part of what attracted me to Hold'em is the idea that it might be unsolvable: there's surely an optimal way to play, that will pay out over a very large number of hands, but so much of the game is based on human psychology which can be wildly variable.

Do you anticipate that AI's can become unbeatable at this game over a certain number of hands? (Are we there already?)

Is there a psychological component to the game that cannot be solved by an AI? (Where a human player, on a shorter run, might do better against an erratic or seemingly illogical opponent?)

brains_vs_ai58 karma

dong :I believe any game should be solvable, its just a matter of can we do it within our lifetime. With computer power/better algorithms, I dont think we are too far off. Fwiw, I dont think Libratus is anywhere near "perfectly solved" in terms of game theory. I just think us humans have been so far from the true equilibrium.

I dont think Fix limit holdem/chess/go is perfectly solved yet either. Its basically solved in terms of humans vs AI

mitchr159828 karma

What's to stop Carnegie Mellon from using the bot to make heaps of money on stars?

brains_vs_ai42 karma

Professor Sandholm told us the AI will not be released to play poker online.

YoungNaijah28 karma

Does Libratus have an understanding of perceived opponent player strength? To be specific, could it rank Dong, Jason, Daniel, and Jimmy?

brains_vs_ai61 karma

Noam: The bot calculates what it thinks they should do in their situation. It can tell if it thinks they are making mistakes.

YoungSchloop24 karma

Do you see this progression in AI technology as an issue for the future of poker, specifically regarding online play?

brains_vs_ai73 karma

Dong: Not in the near future, but we should be worried. I'm no rocket scientist, but I assume that anything with computers grows exponentially. The end is near. It was a good run.

ChickenBake8822 karma

A few for Noam.

1) If you are summing the mirrored hands, it's possible that using equity chops will lead to higher variance in the samples. A simple example being one hand gets all in on the turn and the mirror gets all in on the river. Do you think the equity chops will overall lead to less variance?

2) Have you thought about setting up the experiment where the bot plays against 1 human and the mirrored hands are the bot playing against itself?

3) Any comments on the "DeepStack" (https://arxiv.org/abs/1701.01724) bot which was published recently? How strong / weak would it be compared to Libratus? The paper claims the algo runs on a single desktop machine, although it looks like it was trained on much more hardware.

brains_vs_ai28 karma

Noam: 1) We checked after the last competition and found that equity chops reduce variance even with mirrored hands, though not by much.

2) There are a lot of variance reduction techniques out there, but they can be difficult to verify externally. It's easier for outsiders to understand and trust mirrored hands and equity chops.

3) This competition has kept me super busy, so I haven't read the paper in detail. It looks interesting, and some of the techniques they use are similar to our nested endgame solving approach, but it's impossible to say how it compares to Libratus based on the results in the paper. I would need to see results against benchmark bots, or against top Head's Up No-Limit Texas Hold'em pros in a format like ours. Honestly though, I can't imagine it's stronger than Libratus.

LolPharisees21 karma

I play midstakes 6 max, should I just quit poker in light of this? Can I not just assume that I should expect 3 bot assisted Russians per table within 6-8 months?

brains_vs_ai34 karma

DONG: good thing for you, its much more complex to solve a ring game. Not only is there actions, betsizes cards etc, but now you have to deal with multiple players. I would march forward

brains_vs_ai31 karma

Jason: This A.I. is only for heads-up. The technology just isn't there for 6max yet (although there have been successful bots in the past). I think you are still a few years away from having to worry about that, but be careful.

daaaaaaaaniel19 karma

If you do end up getting beaten by the AI, do you think it will help or hurt getting online poker legalized in the US?

brains_vs_ai80 karma

Jason: I think it has little effect. I think the mechanisms for legalizing poker have more to do with special interests than anything.

noouts14 karma

have you tried triple range merging?

brains_vs_ai21 karma

DONG :Its always a weird feeling to value bet two streets then realize you have the bottom of your range, and its time to bluff. so yes :)

manubfr13 karma

Question for the developers: if you win this, are you staying with Poker? What would your next target be? Moving on to other games or projects?

brains_vs_ai20 karma

Noam: I haven't thought ahead of this goal really. But I think it would be nice to branch out a bit.

oskar66913 karma

Have you tried rubber banding it? Being NLHE, most likely deep stacked (none of the streams load for me) it really has no chance of getting away with cookie cutter GTO play, so it has to react to your tendencies. The question is how long will it stick to the optimal play against your betting frequency and when does it realize it's being duped. So when you 3b excessively pre or cbet 100% it will adjust, right? So when does it stop adjusting and is there a window that lets you exploit it?

brains_vs_ai20 karma

DONG: you are 100% right, it does come back with a better counter and essentially the the exploit is usually gone by the time we can come back the next day. I guess its just a mater of constantly trying something different, but that could also be a costly mistake. I wish we had a HUD...

Now that I think about it, Im pretty sure I couldnt beat the top players that I play on a day to day basis without a hud, let alone this beasty AI bot...

SnakeFarmer12 karma

Hey guys! I'm currently writing my dissertation using poker as a testbed and stumbled upon this whole poker playing computer stuff when my adviser suggested I "build a poker simulator" to test my theories. Haha.

My questions:

  1. Does this bot use the counterfactual regret minimization technique?
  2. What makes this bot better than Claudico?
  3. What made you poker pros even want to do this? Are you guys getting some of that sweet Carnegie money?
  4. Any advice for us academics just now getting our feet wet?
  5. I saw an article online about using AI to identify skin cancer. Obviously, this isn't just about poker. Do you guys see yourself trying to apply these skills to other fields?
  6. Do you guys ever overlap/work with the CPRG in Alberta?
  7. The Cepheus team claimed to have "Solved" HULP. Given that you're playing no limit, how big of a jump do you think that actually is?

brains_vs_ai25 karma

1) Noam: Yes, the bot uses a custom variant of Monte Carlo Counterfactual Regret Minimization, with a form of Regret-Based Pruning mixed in.

2) Noam: Claudico had to use card abstraction to keep the game tractable. That is, it combined similar hands, like a queen-high flush and a king-high flush, and treated them identically. This works in most situations, but against top pros the difference between a queen high flush and a king high flush is pretty important. The inability to distinguish the subtle differences between these hands is primarily why Claudico lost. There were some other factors too. Libratus doesn't use card abstraction. It determines a unique strategy for each situation it is in.

3) Jason: We are getting paid (splitting a prize pool of $200,000), but it's also a really cool experience.

4) Noam: Find a problem that gets you excited. Also, don't get discouraged if things seem impossibly difficult at first. Everyone feels like that for the first 1-2 years.

5) Noam: There are definitely applications beyond poker. None of the techniques we use are specific to poker. They can be applied to negotiations, auctions, security interactions, or any strategic situation where there is hidden information. I also see this as fundamental research into the problem of dealing with uncertainty in the real world.

6) Noam: Yes, I know the CPRG guys at Alberta really well. Our work builds off of each other. CFR was developed at Alberta, for example (though the lead developer of it was also a CMU alum).

arcangel09211 karma

Do you get to see what the computer's holdings are after the fact?

If so, has it helped crack some of the "mindset" that the computer has, or is it pretty balanced?

brains_vs_ai20 karma

Dong: Yes, we get the logs at the end of the day. It is actually pretty entertaining to go over hands you got bluffed on. Its also pretty fun to go over a big bluff you made but realize the bot missed its draw and mucked.

Ckaine811 karma

For the developers and Dong/Jason: What's the computing power necessary for this? Isn't the bot tanking way longer than any reasonable time bank online would grant? If there is a lot of power required for this aren't we a long way off from Joe smith running something like this on their laptop?

brains_vs_ai25 karma

DONG : " Isn't the bot tanking way longer than any reasonable time bank online would grant?"

~Yes, so this is why it cant be applied in real time for online poker. as of yet...

DaveShoelace11 karma

If you were to redo this challenge in say a year, how would you change the conditions of play to give you a better/fairer chance?

brains_vs_ai25 karma

Jason: We'd rather play from home over the course of a year. But then we couldn't have an event like this.

SoundBearier26 karma

Will you still have a home after the bot wins it?

brains_vs_ai15 karma

Jason: I'm not too worried about the AI affecting my living conditions just yet.

ezzyrd10 karma

What do you like most about what you do?

brains_vs_ai21 karma

Dong: I like that I have the flexibility to do what I want from wherever I want. That's not a freedom that I'm ready to give up yet, so hopefully Skynet doesn't get here too soon.

Gorillamath9 karma

  1. Is the bot doing much leading on any streets in single raised, 3b, or 4b pots?

  2. What's the worst hand you've seen the bot call a preflop jam with? Worst hand it has jammed pre with?

  3. Is it going animal in srp's where it's faced with a river bet after action has checked down either as the pfr or the caller?

brains_vs_ai18 karma

Jason: The bot is leading on basically any street in any type of pot. It's quite unusual to deal with.

We saw it get in 66 preflop once, maybe 33 but I don't remember.

Yeah it will go HAM on the river after checks quite often. You get in real tough situations where you never have that strong of a hand.

HoodDuck8 karma

Doug and others in EE have stated many times that you guys don't use solver software for learning HUNL. Without software, how are you able to determine pseudo-optimal frequencies that you can exploit should an opponent deviate? Is equilab truly the only software you use haha?

edit: and PT4 obv

brains_vs_ai20 karma

Jason: We truly do not use any solvers. We bust out the spreadsheets and the equilabs to get the answers we need. You don't need a solver to determine optimal exploitative frequencies if you understand poker theory and math.

Slowta7 karma

With the exception of card removal does the bot even pay much attention to his holdings strength? Seems like he just narrows his opponents range and soul crushes them that way.

Follow up question. Is there a place online I can buy a Cheet poster?

brains_vs_ai11 karma

Dong: I don't think the bot works that way. It's trying to use a strategy to get the best results without necessarily knowing who it's playing. I don't think it looks so much into what we do. I know there's a learning process at night, but I don't know exactly what the methods are. When I found out, I'll leak it out to you all.

NoOneEverPaysMeInGum7 karma

How do your red line and blue lines look verse bot?

brains_vs_ai12 karma

Jason: Overall, we're slightly losing in red line and mostly losing in blue line.

forava76 karma

Have you picked up or noticed any patterns/habits? Does this AI remind you of certain player(s)?

brains_vs_ai17 karma

Jason: I would say a lot of the basis of the AI's style reminds me of how Doug Polk plays. But overall I would say the style is unique.

Libratus mixes it's strategy across all sorts of actions and bet sizes so picking up reads on its ranges is very difficult. We have statistics on the frequencies it does different things, but it's near impossible to come up with anything too concrete.

Meeooowth6 karma

I haven't kept super up to date with totals, but I follow you guys on twitter so have seen a few days results and they're pretty terrifying tbh! I was holding out hope the AI wouldn't be up to taking on competent high stakes regs yet.

Couple questions: 1) What leaks does the bot have? (If any) 2) Is there anything the bot does that is way different than a normal reg? Like, situations all regs would bet 1/2 pot, it checks/overpots/etc. 3) How aggressive is the bot in general? 4) Given the hands you've played so far, what would you guys estimate your long term winrate (or lossrate) is?

Thanks guys, gl with the rest of the challenge. Hope you beat the bot down and we can prevent skynet for a few more years at least.

brains_vs_ai45 karma

Dong: The bot isn't particularly aggressively. But when it is, it is with massive sizes and very difficult situations. So the stats are misleading... they make it look pretty passive. But that's because when it does bet, it bets huge (and also sometimes mixes in small sizes). I think the bot just gets off on that.

brains_vs_ai19 karma

Dong: Biggest difference is that most human poker players do what Libratus does, betting with multiple bet sizes, but humans only have one or two usually. Libratus mixes in a bunch, maybe 10 or 15, and even mixes in situations that I thought didn't make sense before this competition started. It would be way too complicated for a human to do this sort of thing correctly.

apollospaceprogram6 karma

Question for anybody who would like to answer: How did you get into professional poker? What kind of dedication did it take to master the game?

brains_vs_ai6 karma

DONG + JASON :I got into poker right before "black friday", where the major pokersites left the US. I was in college at the time and played poker as a hobby. Speaking for myself, I was obsessed with the game at first, then slowly moved over to the "dedication" mindset. Ive always had love for the game, especially HUNL

FinalFntasy5 karma

I'm coming in to this AMA without much back story of Libratus or how it has been doing against the Pros.. but my question is this:

Isn't the bot just playing percentage based poker? It's gauging your hand based on your previous hands shown, bet sizes, position, etc. etc., and then making the call, fold, or raise based purely on the math? If that's the very dumbed down basics of what we're perceiving, could the pros strategize against it? Throw some mathematically bad calls or bad raises in to the mix to confuse it's data? Thus taking advantage of this thousands of hands later?

Thanks for doing this! I'll be watching the twitch stream today!

brains_vs_ai15 karma

Noam: The bot has learned its strategy by playing itself in poker over trillions of hands. It has tried everything, including "bad calls" (hero calls) and "bad raises" (bluffs), so it's prepared for those. If it could have been exploited in that way, it would have already adjusted its play to prevent that.

Gorillamath5 karma

Developers:

Is there a libratus that knows how to play Pot Limit Omaha?

If not, how much more challenging would it be to create a bot that plays huPLO as well as Libratus plays huNL? Several orders of magnitude?

brains_vs_ai10 karma

Noam: The standard for research is No Limit Texas Hold'em. However, none of the techniques are specific to poker (or Texas Hold'em) so it would be pretty easy to make a Pot Limit Omaha bot using the same approach.

Kriegersaurusrex4 karma

What sort of strategies do you guys talk about at the end of the day with each other to hope to improve your play during future matches? I've followed Jason's stream, and he's said that it's been difficult across the board to land any edge where big wins seem few and far between; and the bot plays unlike any human player rationally would.

brains_vs_ai7 karma

DONG:

A nice list: Frequencies in all situation, bluff to value ratio for every sizing, heat maps on ranges it defends/3bets by sizing. type of hands it uses to make big bluffs (surprisingly, this part is the most inconsistent). To name a few :)

Insomnia13053 karma

What's the difference between Libratus and things like Pokersnowie for example?

brains_vs_ai2 karma

Noam: I don't know anything about Pokersnowie. But there is a lot of brand new research that has gone into this bot so I'd imagine that is all different.

brains_vs_ai2 karma

Jason: I don't know how PokerSnowie works exactly so I can't compare. But we know Libratus is the result of cutting edge learning algorithms that analyze trillions of hands played, and it is the only bot/solver that has done that.

brains_vs_ai1 karma

Libratus has more betsizes. Afaik, i believe it approaches poker in a similar way

weegbeeg3 karma

At what point did you realize this bot was not like the previous bot?

brains_vs_ai10 karma

Jason: When it stopped limping.

ZondiaC3 karma

Hi guys, Who decides if you, specifically, are able to represent humans? Do you think a group of other poker pros could potentially beat the AI? Thanks

brains_vs_ai3 karma

Jason: We are not the absolute top 4 players, so if you got a lineup with Doug, sauce, etc. they would have a better shot. I would really like to see how they would do, but I think it would be a real tough time for them too.

ABoss3 karma

Dong what kind of dog is Bambi?

brains_vs_ai5 karma

Dong: It's apparently a deer of some sort.

rockyrosy2 karma

Hey guys.

Do you think the bot has come in with a preloaded close to GTO strategy, or do you think it's adjusted it's play as the game's progressed?

brains_vs_ai7 karma

Dong: I believe it's the latter. I don't think it's a perfect GTO strategy. I think it's trying to accomplish a strategy that's closest to optimal versus us in particular. There are a lot of things that have been happening... For example, Libratus never used to value bet for 3 streets (bet, bet, bet) in any action. But now I see that more often, and it's because we've all been calling down on the wider side. It's just a hunch... only the creators know.

Noam: We'll tell them all the details after the competition.

ImSrslySirius1 karma

DOES THE BOT BLUFF???

brains_vs_ai8 karma

Jason: The bot does indeed bluff.

Noam: No, it just counts cards.

Deegsta1 karma

In a friendly game we had an issue come up after final betting on the river: the caller showed his hand before the original raiser showed his and the original raiser mucked his cards declaring he didn't have to show because he was conceding the hand to the caller. Must the original raiser show his hand?

brains_vs_ai2 karma

I dont think so if it wasnt an "official" SD to see who has the best holding

Bstings6241 karma

Dong: what does BDDK stand for?

brains_vs_ai2 karma

"being dare devil kim"

isuckwithbikes1 karma

You have been playing some really long days, have you started dreaming about poker?

Were they good dreams or bad / what happened in them?

brains_vs_ai3 karma

Jason: Lol not quite yet but yeah it has been quite the grind. Most of my dreams are just about sleeping longer.

Zer0Summoner1 karma

Have you been playing the machine normal like you would a human, or have you been trying to infer it's algorithms and try to disrupt or confuse those?

brains_vs_ai5 karma

Dong: At first we treated it like Claudico (it's predecessor) and that was a huge mistake. We tried a few other things after that which didn't work. Then we tried playing it like a human, and it wasn't working well but not awful either. So we shifted strategies and tried to find some other weaknesses, which hasn't been going great.

tn_notahick1 karma

Are you learning how to act like the bot, so that it will make you a better player against real people? (Is the bot teaching you? )

brains_vs_ai1 karma

Jason: Playing against a tough/better opponent will always make you better because it forces you to play each part of your game the absolute best possible. You have to fight for everything. That practice makes me better.

chunks2021 karma

Question for the players: what kind of effect is this grind having on your mental state? It seems grueling.

Also, how are you enjoying Pittsburgh? Have you enjoyed your free time here?

brains_vs_ai2 karma

We had no free time in Pittsburgh. I flew in at 10pm and got to the casino by 930am the first day. Been doing something similar since.

Gorillamath1 karma

  1. Will the bot's stats be published after the challenge?

  2. What other hunl players were considered for this / what was the selection process and who decided on the final 4?

  3. Is Jason this much worse than the other 3 or has there just been big variance with spots (i.e. The bot 3b 83cc vs dong's 97cc, 1.5x'd the A87r, dong called, and it checked down. Daniel flatted 83cc vs the bot's 97cc, xc'd a .25x flop bet and it checked down)

brains_vs_ai6 karma

1) Noam: We send the hand histories to the pros, but they asked that they not be released publicly.

2) We invited all the pros from last time (Doug Polk and Bjorn Li), but the scheduling didn't work out for everyone. Dan and Jimmy came recommended from the other pros.

3) Noam: We can do a calculation afterward to see how lucky/unlucky everyone got. I think Jason has definitely had a bad run the past few days though.

noouts1 karma

have any individual hands stuck with you?

or looking back over the logs any hands you're particularly proud/ashamed of?

brains_vs_ai2 karma

DONG: its all such a blurrr. yeah there are hands but I might get the hh wrong. Ill try to touch on this once the challenge is finished

TenchiRyokoMuyo1 karma

This is for Noam and Dr. Sandholm, or really anyone that's been playing against the AI:

Is the AI playing statistically only, or is it building a psychological profile of those it plays against, and playing according to that?

brains_vs_ai1 karma

Noam: Sadly, that's one of those things we can't answer until after the competition.

noouts1 karma

also curious about the bots all in pf range? any interesting pf shoves/calls that stand out?

also has it infiltrated your dreams/nightmares yet

brains_vs_ai2 karma

AI raises 85% from the SB. BB varies by our raise size. I sleep like a baby

emilykiss1 karma

How fast is the bot usually? Would it work in a not head's-up situation? Is it that much better from the other bots doing 20BBs on a full table? Are you guys feeling tired and do you feel like you lose hands because you guys are tired, but the ai can't be?

Also, this is not meant to be something bad, I just haven't done a thorough search, are you guys "top professional poker players" because you guys won big tournaments?

brains_vs_ai1 karma

Jason: The bot takes about 45 seconds to recalculate its strategy on the turn. Sometimes multiple times, so this would time-out on poker site's timelimits. Obviously this will be solved over time as computing speed increases and becomes cheaper.

Yes, we are tired and it causes us to play worse but I don't think it's the difference between us winning and losing here.

brains_vs_ai3 karma

Jason: We were selected because of our skill in this specific poker variant, heads-up no-limit. Tournaments are a whole different game. Heads-up is much more specialized skill, and niche community. Not in the mainstream.

BelatheDog1 karma

Question for Jason - Is your shirt itchy at all?

brains_vs_ai2 karma

Jason: Nah brah

sash1871 karma

Are you guys supernovas?

brains_vs_ai1 karma

super nova playing HUNL is basically impossible. Its a good thing, since there is a bigger edge/lower rake

Redditer10001 karma

If you could see the bots holding after every hand do you think you could beat it? Or is it just that balanced?

brains_vs_ai2 karma

DONG: I dont think so. The IGT Limit holdem slot machine shows its holdings. It would help. Being human, I think i would make bigger folds if its close since there is instant gratification to satisfy my curiosity.

dracarysmafu1 karma

Where do i start to become a professional poker player like you guys?

brains_vs_ai3 karma

Jason: Just play poker for fun, and play really small compared to how much money you have. If you have a passion for the game and want to be good, it will grow. Use resources like UpswingPoker.com to improve your game. You won't make the conscious decision to be a professional, it will just happen.

masterpps340 karma

Question for Everyone, Jason recently said on stream 'I'm not to worried about the AI affecting my living conditions yet..'

My question is, what if it does? What if it gets in your head so much that its always 'in your head'. You can't sleep right anymore, you can't eat right, everything seems upside down. One second you're laying down in your living room trying to get some well deserved R&R, the next second the AI's code has intentionally embedded itself within your subconscious and is eating away at your life force..

How will you overcome this very real potential?

And what would be your response if the AI was for lack of better words, to 'take your girl'?

brains_vs_ai2 karma

Jason: If this happens I will be forced to never leave the VR world in my HTC-Vive. I will permanently be a space pirate trainer.

Boshimonos-1 karma

Can you have Dan Bilzerian play the AI bot? He took some crazy lines in his heads up match so maybe that will work better? Maybe he would give the computer some of it's own medicine? (Yes I'm aware Dan is worse than anyone playing against it right now but I'm curious if he could at least beat the AI in the short term)

brains_vs_ai2 karma

Dong: I would love to see Dan Bilzarian play this.

brains_vs_ai1 karma

Jason: Would for sure be fun to watch. However, the A.I. has a significant cost to operate on an hourly basis so someone would have to raise the money to arrange that.