If you had played 3,000 hands, then that would not be enough to know your win rate against our AI. You really need to play 50,000 or more.

Yes, our AI learns. So when you first play and you are reraising a lot, our AI will not know you are reraising a lot (i.e., it takes it a while to know how much you are raising). However, by now it should be responding to you.

When in the blinds, most pros (and our AI) recommend folding the vast majority of the time since you'll be in bad position throughout the hand.

Most pros recommend folding to a reraise most of the time as well.

Anyway, our AI's biggest weakness is preflop 3 and 4 bet game, so I'm sure if you are focusing on that you can find ways to exploit it a bit. Playing against a full table of our AI, it's likely much easier to find ways to exploit it, then in the real world (where you usually aren't playing against a full ring of the exact same player).

The greatest learning opportunity with our software is your postflop game. I would encourage you to import a bunch of hands, and look at your biggest improvement opportunities. You can review hands and learn at an extraordinary rate.

I play exclusively HUSNG can I use this?

It does not play Heads Up at this time. Right now it is just for 6max and full ring cash games.

We hope to add HU in the future, but I do not know when it will be completed at this time.

Of course improving your 6max and full ring game would likely improve your skills at any game of poker; however, it would be indirect.

Ok.

I have now played against the toughest AI 9 handed FR 20 times

Each and every time after 100 hands I have the most chips.

I do think it's a bit too passive in terms of steals.

And you said I need to play 50k hands to know if I am winning. But how many hands does it need to play better against me? I can't see how it will beat 100NL. I can't beat 100NL but I am crushing ur game. And believe me I wish I wasn't for then I could learn. And boy do I want to learn and improve. I don't believe this product is good for 100NL players. I think you have targeted the wrong group when you say this game can beat 100NL. And I seriously doubt it can.

I am not calling you guys a liar.

And I am not saying it isn’t a great product.

I just think it needs to be aimed at more complete beginners.

Sorry that’s not fair, I don’t mean just complete beginners. But I don’t think 100NL players who break even or better will get much from this. I guess anyone doing worse, or playing smaller games will.

And I love the idea, and even the product, just not for 100NL players.

Some notes on the product (i.e. some things that may be bugs):

a) on the play game if you stack someone the money isn't correct, I can't work out what's going wrong but you end up with too much cash

b) it's quicker to play at a slower speed, if you want to fold a hand, and skip to the end, if you have it set on the top speed (10) you won't be able to click 'skip' before it disappears. So it's quicker to play at a speed of 5. I think maybe 'skip' needs to be permanently up to click. So, when you get a junk hand, you can 'Rush' like fast fold and get to the next hand asap

c) the advice is WRONG sometimes, I mean obviously wrong. I don't mean I disagree, I mean your program will disagree with the advice. I've been told to fold QQ on the button to a raise. Then when I replay it the advice become correct and it's not a fold. I know this is a bug, but as the advice at the time is what a beginner will use/learn it needs to be fixed asap. I have noticed this a few times (3 times in my last 50 hands alone), but have only copied one instance (the last one) an AK hand, if I can put the pictures into this post I will. I have during hand advice and the correct after (replay)hand advice. I hope this will be at the bottom of this post.

d) Sometimes you don't get told what you did wrong, you just get a $ amount in the Deviation box. It's obviously 'bad play' but it just says something like '$4.14'

e) When you use it live (real money), it could be more helpful. By that I mean I noticed it was mirroring my bet sizes. So if it agrees with me to bet, when I replay the hand it says to bet the amount I did. I couldn’t believe it was totally agreeing with me, as I was varying my bet size depending on situation and opponent, some of which your AI couldn’t know. So I started to vary them bigger. And still it kept saying after the hand that I should bet the amount I did. Now this isn’t the best way to run it. Because if a newer player is betting too much, or too little, he should see what is a more optimum bet size. Which the player can notice and learn more from. So rather than conforming to the players bet size, the replayer should be saying how much it would bet. So it tells the player that he did the correct action, and the AI would bet this amount. That’s not to say that a bigger or smaller bet is wrong, but to have a good guide will help newer players who aren’t sure. It’s not always enough to know that you should bet.

f) back to the play table, I’m used to it, I run it full screen on a 50” Monitor. I had a new player, a friend who has NEVER played live before. He took ages to work out actions. Ok he’s new, but (this is full screen 50”) I think the table could be cleared up a bit. People’s bets closer to them, and maybe bigger?

I will post more when I get it.

My trial ends in about 7 days.

If you want me to beta test it for you, for longer let me know how to extend the trial period.

Like I said it is a good product but not for 100NL (in my opinion) and already I have suggested it to a brand new player. As I think playing the Play (offline) section will teach him so much, without having to lose a fortune learning it.

And the other point, for you, is if players play on poker site ‘play’ money games, they are NOT learning what the real money games are like/play like. Whereas with your software they can learn this. They can play for play, but against ‘real’ type opponents.

Ok, that’s my report so far. Now let’s hope I can get those photo’s up.

PLAY GAME

DURING <wrong> advice:



AFTER <correct> advice:

Not sure if the above worked

Thank you for the report. Here's what's happening with those two hands. In that situation we are simulating the result and the differing advice is due to variance. Postflop we are providing EV's so that people can see a call is for example a call w/ an expected value of $0.20, then if it's reanalyzed and says fold for a negative $0.10 the customer can see that it is variance. Preflop we don't report that, but we'll fix that soon. Playing against two raises or more is simulated. You can also go to "help" and select more simulations to reduce variance if you like (at least on the new version).

We had a bug that the customer was getting paid too much in all-ins, but we thought we fixed it. Can you tell me what version you are using? You may need to download again from our web site to get the fix.

Also, if you have the early version then it won't respond to your excessive preflop raises. The new version does.

Both of those issues could explain some of your performance.

BTW... please import hands and look at your biggest deviations in the reports. Several players of your skill level (and even a SNG poker coach) have made terrific improvement in their game in a short period of time.

Also, remember to go to www.holygrailofpoker.com and make sure you have the most recent version of the software. Hopefully that is the fix for some of your issues.

d) you can see what you did wrong in the hand history it'll say something like

check (B, $4.15) this means you checked, but you should bet for $4.15 more in expected value.

e) Yes, if you bet $100, then we evaluate a bet of $100. If you check advice before you act, it'll recommend what it thinks is best, If you act then look at advice it will compare what you did to the alternative.

f) Yes, we've gotten feedback that we can improve the graphics in this manner. We'll make some big improvements here. By resizing the screen to the ideal size for your monitor will help a lot as well.

Hi I realise that but I was talking about playing for REAL money, if you read it again.

So it won't give you advice BEFORE you act

Hi

I have whatever version was on ur site about 8 days ago, so I guess the most recent.

Re the reload.

Just did an experiement.

Set everyones stack to $10 (play 5/10)

Now when anyone re buys they go to $1,000

Not sure if that's the bug that I get when playing normally but the stacks aren't correct after an all in

Right, it's set to rebuy to full stack. We should probably set it to rebuy at whatever it was.

Paul,

I've been playing around with this some more, and I find it very helpful, as a relative newbie to NL. I also have a question/suggestion. When you are replaying an analyzed hands, there is a sidebar section called "Hand probabilities," which I like. However, I can't find any way to view the assumed hand range for villain that these are based on. Is there any way to do this? If not, there should be.

Thanks.

Yep. This is a great idea. It's in the plans, but we have not determined the ideal way to display it. Here's the challenge... we have a probability of all 1326 possible hands. Sometimes (e.g., one of each suit on the turn), there are many fewer hands.

My guess is that we will show something like this (if board is Ks 7h 7c)...

7d7s - KhKc 2%
Ac7d - 7d2s 3%
AcKc - Kd2s 7%
etc...

However sometimes it would be cool to know...
AKo 8%

Also, it's a lot of data to store in the db for EVERY decision of every hand. So maybe we have a button that you can push and it determines it and write it to the screen (which would take almost no processing), but after you leave that hand then next time you wish to see it you must press that button again.

We could also show the AI's assumptions of the way the opponent will play every hand in every situation during the hand (which is what creates the probabilities). This is a lot of data which might be interesting if someone wishes to really do some in-depth analysis of a particular hand.

Glad you are enjoying it!

Paul, since I know you are open to constructive criticism, I will say that the above is not the way you would want to do this.

A better option would be a pokerstove style display, with possible holdings shaded on the standard preflop grid. You could use lighter and darker shades for weighted hands as well. At the very least, if you wanted to make a list of hand types as above, I think you'd want to remove individual suit info from the cards.

I think this is the right way to do it.

Yes. This is another feature I'd really like to see. In particular, a lot of the times the advice will say, eg, "Check", and I think: "Check with intention of raising, calling, or folding?" Sometimes the answer is obvious but many times it's not, and in those cases the simple action advice is not worth much.

Hope that helps.

I like the idea...

So it would be easy to show the 169 hands (in standard 13x13 matrix) w/ shading. Then perhaps when the user clicks on or mouses over a hand (like AKs) he sees...

AsKs 1.1% AcKc 0.1% AhKh 0.1% AdKd 0.1%

What do you think?

Yep, exactly what I was imagining.

Do you plan to support merge / rpm ?

We plan to support Merge, but it'll be at least a few months. I'm not familiar w/ rpm. What's that?

A merge skin. It shouldn´t be too difficult to import other hand historys.. Thx for your reply.

How fast the hh importing is supposed to be on average? I have tried a few times to import some hands and it just keeps going for hours and eating cpu until I finally need to use the computer for something else and have to stop it.

It analyzes each decision and each decision takes 1-2 seconds. We are trying to speed it up, but it will always be dramatically slower than imports that don't provide decision analysis.

So, I would just do what you are doing (i.e., import when you aren't using your computer).

You can also select a faster analysis on the help page if you like -- this will speed it up.

This is fixed in the next release. Opponents will re-buy to whatever the user set.

played 4 6 max tables of your "optimal" bots as well as some mixed ones.

weaknesses: 3bet/4bet game (very weak), not adjusting anywhere near as quickly as a human, donkbetting range (donkbet/fold range), late street aggression, weird taking back the lead on the turn lines...

strengths: barreling (sometimes), pot control, value betting...

its an AI that can beat fish and robotic players with leaks imo. maybe it would do better in rush poker where short term game-flow dynamics largely disappear.

Basically, there would be a line of small and micro stakes guys quite happy to play nothing but a table of your bots for a living. I can't think of a more profitable read than "I'm playing a table of robots". I can't say the same for "I'm playing a table of people who watch poker training videos".

Thank you for the post. True. It would be nice to be able to play a table of all the exact same AI (or exact same person I guess). You'd get 5-9x the read on them that they would get of you. Moreover, humans can adapt to changes in opponent behavior much more quickly than computers -- no question! I think the people who could make a living playing our AI could also make a living playing low stakes as well. (However, I guess I could also imagine someone simply becoming an expert at exploiting our AI.)

You are also right that our AI is not very exploitive. It plays solid poker and benefits from opponent errors. It's very hard to make highly exploitive AI that isn't in turn also quite exploitable. And, again you are right -- it would be better at rush than regular for the reasons you mention. Currently, it does not have a concept of table image. Said differently, it treats an opponent who raises 10% of the time the same as an opponent who raises 10% of the time including the last five hands.

Based on your comments I suspect you are a very good poker player. To me, the greatest value of the software is in identifying postflop improvement opportunities. You might try importing some hands and take a look at some of the deviations. I'd be curious what you think of some of the hands that the AI thought you could have played better.

Some very solid poker players have plugged some material leaks in their game with this approach. Combining this with some of our videos is also powerful -- you can compare how you play to both our AI and to two of the all time best $200 buy in NL players (i.e., how you play fundamentally not just in terms of aggregate statistics).

We have released a new version: 1.1.117

You can download it from our web site at www.holygrailofpoker.com -- just click on Step 2: Download Free Trial.
You can see the list of new enhancements and bug fixes here.

As always if there is anything we can do to make the software better for you, please don't hesitate to share it with us on our forum.