Ben, 

Very cool video concept. I was excited to watch the footage. The biggest thing I took from the video was delight because it seems that the bots have a long way to go to beating a good NL player.

3 years from now when poker experts credit ben sulsky for revolutionizing bet sizing with his never check / at least a minbet strategy, he'll rightly give credit where it's due

really interesting video!

interesting but IMO would much rather see you reviewing your own play (or other good humans) than making more of these bot vids... probably too late anyways so w/e

Have to agree with WhiteDevil. 

Both of these bots seem to live in their own universe with the very weird plays that they have shown and the ambiguity of not knowing how they arrived at their decisions / how they were programmed in addition to the very small sample size (they could easily play 100k+ hands against each other and we could review them in HEM) makes it really hard for me to draw any valuable information out of this. 

I'd also much rather see a human play one of these bots over a large sample than 2 bots each other. We don't even know if they are capable of adjusting strategies. 

So yeah, kinda nice for a gimmicky video but please don't make this a 3parter (aka 1,5 months of content).

I third this notion. It is undersandable that he wouldnt want to review his own HU play in too much detail though. Rather, maybe it would be nice to learn a bit more about how to set up proper ranges for HU play.. and how to balance those ranges. Maybe tell us what you think the bots ranges should be?

I found it at least somewhat interesting.  Would I prefer it to an actual Sauce match, no.  But there's usually some value is seeing "outside of the box" play, which I think we can all agree we saw here.  I particularly enjoyed the min bet strategy the bots were employing.  Maybe betting 1x into 30x isn't good, but it made me think about how using multiple bet sizes as part of an optimal strategy is almost certainly correct.  It seems so likely there are a lot of spots where certain hands in our range make more by betting 1/3-1/4 pt rather than just 2/3'ing everything.  It was a lot more stimulating to watch than standard hunl videos where all you see is, "I 2x pre.  I bet 2/3 pot on the flop.  I bet 2/3 pot on the turn..."

If I could recommend a future video, I'd like to see something about the theory behind splitting ranges preflop in HUNL by playing limp/3x and ways that we could balance those ranges.  Obviously Ben, don't get into detail about your specific strategy, but maybe sprinkle the theory behind this concept in as an intro to a hunl video.  You could even touch on it in this series as the bots are using many different preflop sizings.

This was interesting to watch but just as few people above mentioned, I would prefer this to be just 1 video, not 3 parts. I would much rather see you continue doing 5/10 NL series which so far has been very insightful and awesome. 

I'll disagree with what others have been saying, I think this was interesting enough that I'd be happy to see multiple parts.

So far every small bet has been a weak-medium hand, which would seem to be easily exploitable if they don't protect those ranges with some strong hands. I wonder if we'll see them doing that at any point.

More videos like these.

Ben, Im signed up to this site because of you, and being able to watch Phil's back catalogue. 

Making this a 3 part series would make me pause the subscription. It's not that the bots aren't interesting, it's more that the disclaimers you yourself made at the start of the video make the information that can be drawn out of this pretty non applicable to our own poker games. 

I would have expected a video producer at run it once to at the very least have watched the entire footage beforehand and condensed the findings into either a speedier live video or more of a powerpoint style presentation. 

I'm looking forward to future videos, and I hope the message comes across the right way.

I have to disagree too. I really enjoyed the video. And helps me to confirm some "standard" decisions that i was taking, and see that im going for the right way.

I really enjoyed the video.  Interesting analysis by Sauce.  Additionally, strategy video's with both player's cards revealed provide a unique perspective that you can't get playing.

Does anyone know anything about how these bots are programmed/how they work? For example, does a bot "know" all the hands it can have in a certain situation and with what frequencies, so it knows exactly which hands it needs to defend to be unexploitable? Do bots read hands/put their opponent on a hand, or do they just concern themselves with their own ranges?

The Jx9h is peculiar. Could the bots reasoning for small check-raise is that he assumes stronger flushes would've bet very often on the turn? (Not saying it would justify the 2,5x check-raise on the river, but the bot must have some sort of logical reasoning for making a play, whether the reason itself is correct or not.)

As far as the video goes, I don't think there's much merit to putting any more of these bot vids. If the bots had some cutting edge approach to the game which could be incorporated to human players game, then sure. But it seems like bots are way behind the curve atm.

And when the bots are behind the standard player curve, I think it's just bad business in the long run that players as good as Sauce are reviewing bots' leaks in the public, thus contributing to the botting problem and development of better bots.

If you continue with the series, then please explain what would be your standard approach with the hand in progress instead of only commenting bots play. That way it can be an ok learning tool, I guess.

I would really appreciate more demanding(=challenging) content. First part of Toygaming was also too tame. I'd assume almost everyone paying $100/mo for the subscription to have done their homework on basic GTO play, so it should focus more on advanced stuff imo. If not, they can always catch up/read MoP.

FWIW, it seems bizarre that the best HU bots don't seem tough at all. Yet there has been a botting epidemy at iPoker/Party NL 6-max 1/2-5/10 IIRC. Considering that 6-max is an unsolvable (or multiple GTO) game it seems weird to have bots winning there. Maybe the people submitting bots to these competitions (slumbot, Hyperboarean) are only building bots because of scientific reasons, and the people who are doing it for financial incentives are keeping as low profile as possible?

This video was not bad at all but I feel its a waste that Sauce make this into a three part serie. Iam sure most instructors could do a good analyse on the bots, we just dont need Ben for it. The 5-10 zoom videos are in a completely different league compared to this and in class of their own. I feel Sauce want to bring us some different type of videos and that is cool but three videos of this is to much when the material from Ben is very limited

I agree with stepitup...Would be much better if Sauce made other kind of video.

Come n Ben, I would also prefer if u could give us more of your NL 5/10 play. This is not worth your time IMO.

All this hate for Ben trying to deliver a very new style of video is unfounded. Even if you found the video subpar it still is an interesting subject and one that is both pertinent to learning and to gauging the health of online poker longevity in some ways.

Can you watch the whole botmatch somewhere ?

Bens take on nonstandard plays sound pretty awesome to me, but I´d also love to see a LIVE 2-4 tabling match vs a reg at small to mid stakes, as you get a better feel for the flow in those and sometimes it's nice to brush up on 'standard' plays as well.

It´s obviously hard to go in depth on commentary at 500+ hands an hour, but people can ask questions in the comments as well as pause and go back in the videos.

http://www.youtube.com/watch?v=PlfEfKu5tZA .

Given how much academic thought and calculation time (years of both I'm sure) has gone into both of these bot strategies, it only makes sense to assume they play reasonably well and adjust judgement from there.  This is what makes this video great imo -- that's an excellent way to learn from all those hours that the bot builders have put in.  The bots make exactly zero plays without reason I'm sure, and there's going to be some sort of logic and tons of calculations behind that reason.  Understanding and evaluating that reason, that's got to be how to learn here.

What could be done to enhance this series is to prepare ahead of time.  Look through the hands, find the weird spots.  Then look for explanations on why the bots come up with the weird plays they do.  Read papers, ask in the 2+2 thread or even contact the bot builders themselves.  Imo, that's how an initiative like this would transform from something you'd do for yourself, into being a video series that many would enjoy to watch and learn from.

Basically, answering the "why" -- that's often how people learn.  (bots need no why though)

Definitely hoping to see this series continue.

Absolutely loved it

I don't agree that we should assume these bots are good, Sauce. These are not individuals that have been playing poker for years and years, no, they are barely programs that are doing what some HUMAN students at some university told them to do. Humans that probably have little to no achievements in the real poker world. We know very little about these bots which makes watching these matches and trying to rationalize good play from weird play very dangerous. 

Once again, the sample is very tiny and both players are bots. I have not found any way to play against these bots which would make for a much more interesting video. 

@Zachary Freeman: A lot of people (including me) that are complaining appreciate this as a once-off video but just don't want it to turn into a series of 3 videos because for us watching real play is much more educational. 

How much actual discussion about the bot play has been going on in this thread? Very little, people are either stating that they like it or dislike it. Same goes for his Toygaming video which had very little comments overall. But all of his live play / review videos get a TON of comments and people are really engaged. 

this is no other than donk on donk violation

I think this video series is a good lens for viewers to see a Pro through.  Even if you pretend that these bots are regulars from a foreign land we learn how Ben analyzes players' strategies finding flaws, strengths, and giving a comparison to 'Standard' plays amongst the HS community.


The lesson learned from this video was not how good are bots because that was not revealed from this sample.  

The valuable lesson is to open our eyes up to new lines that may be more exploitative and +EV that traditional lines.  According to TOP the further you opponent is playing from how they would play given knowledge of your hand the better the 'edge' or player.  As players are able to hand read on multiple levels better and better the challenge becomes keeping them confused and making high variance guesses [I see more of a passivity trend when my opponents are confused].

I'm indifferent about the release of this series.  I am believe that RIO is mainstream and possibly being pottripped by e-pirates so anything preached becomes 'Standard' therefore I prefer if I don't educate my opponents about macro-strategy adjustments or improvements.  Being a live player is my study ethic and attention to details that create a wider void between me and the 'Regz' and this information would only widen the gap!

I loved this video and would like to see more.  

You said that folding J6o and K2o was standard folds to a 3x raise.  I have played HU NLH in a very long time but I would think these two hand are profitable to defend to a 3x raise.  They play poorly postflop but are still int he top 60% and 50% of hands respectively.  Do you think defending the BB less than 50% to a 3x raise is correct in NLH?

When slumbot had Ac8c and hyperion had Kh3s at minute 40.  The flop was check min bet and called.  The turn paired the 3 and slumbot check folded to a pot sized bet and you said that makes lots of sense.  I was a bit suprised by this fold, I would have thought Ac8c would be high enough in our range to check call after the previous acton of min raise preflop and min bet flop.  Won't slumbot be over folding this turn to a pot sized bet if he only continues with his pairs and stronger semibluffs?  Or are the RIO so bad that we can't call the turn with roughly 40% hot and cold equity vs a range of J,4,3,22-AA,*d*d,56,52,A2,67.

I don't know if it's better or worse but those bots seems to use "inducing" (very small ) bet sizings in many spots

while most of the players nowadays are using close to pot sizing (little bit under or over) to rep polarized ranges

this is two different strategies which have both positive and negative impacts, but I don't know which one is best

if mixing it up is better.

The 99 vs 86 minbet hand is crazy to me. I can't see how minbetting in position is optimal. OOP it makes way more sense.

:D zoo much fun vid, idk how ppl can be not interested in seeing and analyzing these bots 

Is there a possibility you get to play couple of hours in person with these 2 bots and we get to see it?

- why are both bots reloading to start chips every hand, so we can't see which 1 is better then other after lets say 1 k hands, would 1 zero?

-is it actually illegal to use bot for playing instead of u, or its just opinion they should be illegal ?

Anyone else find it wierd that the majority of the time the bots are making 1-3bb bets and minraises, and then slumbot comes out of the woodwork with 88 on KQxxccc and bombs a turn bet with literally 1 out to a river that can valuebet?

Seems pretty lol to me.

this vid is amazing, more pls

With a slightly bigger sample it will be interesting to see how they play big pots.

ye i was thinking ~ they played small pots- what if 1 bot started raising whatever 2 cards whole stack on rvr - would other be folding all but nuts?

I really loved this video Ben.  I don't think the bots plays were bad at all.  It seemed to me the slumbot was slightly stronger but this was over a small sample size.  I'd love to see more content like this and it's made me think a lot of how utilize more bet sizing strategies in my own game instead do the standard 2/3 to 3/4 pot or min raising the btn strategy.  I'd like to see more content like this and get a better understanding of how the bots are splitting their ranges with their bet sizing.  

Interesting watch, and it's great to see that bots have not developed to a point where they can start beating good human players. I had heard that bots could beat 5/10 HU NL games. Well, I was informed incorrectly.

What surprised me most about these bots was their inconsistency. You'd expect them to set up in a way to at least have standard opening/betting sizes here. They were kind of all over the place.

Though it was a good video, I do have to agree that the most valuable videos for improving poker are the ones like Bens 5/10 Zoom videos, Phil's Kanu HU series, the WCOOP HU video, live Zoom videos etc.

Someone in the Chatter suggested a format where a couple of pros, and perhaps some forum members, play a cash game session against each other, with commentary being provides from multiple angles. Poker VT did one of those a couple of years ago, and it was fantastic.

Really, really cool vid. Also one of the funniest I've watched, it was hilarious to see you constantly being flabbergasted by their plays. Definitely continue with this. 

I don't understand all the love for this video. 

I agree that it's interesting to analyze how top poker bots are playing, but this video contain almost no analysis from Ben, it's mostly observations that we all could of done on our own by watching the footage.

This topic would of been better served by creating a thread in the forums with a link to the footage. Otherwise, Ben should of watched and analyzed the whole match before trying to make a series out of it.

I feel like Ben is simply trying to maximize his own time by making a video series out of this footage instead of trying to maximize the value that viewers can get out of his videos. I'm not saying it's the case, but it's how I feel about this video.

I have watched many hands of hyperborean_iro and as expected its use of the min bet is very balanced as the strategy was arrived at via equilibrium finding algorithms. It is making that play because in the game abstraction (approximation) it is attempting to solve it is the most profitable way to play the hand, which of course doesn't mean it is the best play in the complete game. I made another video that illustrates its use of min bets (contains only hands where such bets occured): http://www.youtube.com/watch?v=PlfEfKu5tZA

It is interesting to note that these bots have different styles: hyperborean_iro is much more aggressive preflop and postflop than slumbot so most of its winnings are non-SD as you can see here http://i.imgur.com/Y3e3k5s.png. The last 130k hands are against tartanian6 which is a bit more aggro postflop so the red line starts to flatten a bit.

really enjoyed it ben.  first of all it was something new and interesting.  i especially appreciated when you took shots at forming your own assumptions about their decisions and/or what they might be trying to accomplish.

I meant the play was balanced in the sense that it hides the information well.   Hyperborean can showdown the nuts or a thin value hand on the river when it minbets.  I think that if a play is balanced in the abstraction it won't necessarily be in the real game.  For example, in the abstraction they will typically merge boards with similar strategic properties (to make the game tree size manageable) so there will be information loss and this may lead to less than optimal plays in the real game.

It's hard to make sense of all the things it does. It's like a black box, we can't follow the process that leads to a certain play as it is the result of millions of iterations of the equilibrium finding process.

Very interesting video...as far as the 99 min bet on the 4568ccc board:  went 3x to 15x...so Hyper could be assuming a super polarized range here...it's possible Slum would check a 456cc flop with QQ+ for pot control on the hands Hyper would call with...but would Hyper call a 15x 3bet with hands that crush that flop other than 44-66 maybe 45s 56s 67s 78s cuz they're 200bbs deep?  so like 24combos of pps if u want to count 7s as crushing the flop and all the combos of scs but less because of card removal...all I can think is that it was trying to realize its equity against combos of AJ+ and possibly pot control against big pairs from getting in value bets on a brick river... and when it 4 flushes the river with str8 on board (45687cccc), the check behind definitely makes sense to me b/c of the J9h that Slum had earlier where it checked 3 streets and c/r river for small value earlier but honestly ya idk...super interesting video.

this is so painful to watch...i couldnt make it through the video...if these bots played any human above 100NL h/u they would surely get destroyed

Really nice video! Its a kind of relief to see that the bots still very weak.

Hehe 5 years later and we see just how far the bots were and how bad us humans were.