You're welcome. I'm glad you are taking the security seriously.

Ok so what about adding this practice mode where you can set the cards/board to your liking, but not offer the advice? Just let the user play it over and over as many times as he pleases, and just log the number of hands and chips won/lost and print out a report.

Also, configurable bet pot script is needed.

From what I understand, poker sites already allow certain programs that give advice. What is the difference between allowing the advice given by those programs and allowing advice from this one? Does it have something to do with the quality of the advice given (low-grade simple advice is allowed), or maybe it is the method used to obtain that advice (simple logic okay, complex simulations not allowed)?

In any event, I really liked the idea of being able to adjust the parameters of a hand. I would like to not only adjust cards and actions in a hand, but also adjust the stats of opponents to see how that changes the advice. I think there are steps that could be taken to prevent the advice from being accessed to play a hand in progress.

This seems like it could work.

How would the "bet pot" work?

We'll see if we can come up with some means to adding that capability while preventing improper use. The only idea I've had so far is to force 10-20 seconds between the first action to create a new hand situation and the time before we allow an analysis of that hand.

Or (if possible without your framework) you could measure the EV's variance between simulation roll-outs and either warn the user (in a similar way to SNGWIZ does when it prunes an important 4-way branch...) or even let it run until some acceptable confidence limits are attained.

Juk

@GalacticRewind: No poker sites allow AI advisors that make your decisions for you, leaving you to simply click the appropriate button. You are probably thinking of software that displays various odds and such in real time. Here, you are still processing the information to make the final decision.

I was playing at a table with all best AI players, and it recommended that I steal from the button with Q8o. Isn't that sort of a weak hand to steal with?

Also, I imported some hands, and it recommended that I fold Q9o. So I steal with Q8o, but fold Q9o?

The only significant thing I see between the hands is that for the second one, the big blind had a stack of 80bb and small blind had 160bb, instead of 100bb and 100bb, but I do not think that should make a difference.

There is also the fact that in the second hand, it was imported from a game at really small stakes, but again does not seem important.

So just to be 100% clear, you are denying that you hired people to bot for you on Full Tilt, and that you were caught and all of your accounts frozen about 2 years ago?

Yes, that's a weak hand to steal with (i.e., one of the weakest); however, it recommended that because the two players in the blinds fold to steal attempt a high % of the time (i.e., higher than they did when it said to fold Q9o).

I suspect you saw the link on our forum to some guys claiming I botted. I’d like to keep this thread to support for our software beta launch. However, I’ll answer your question (which is a fair one) and hopefully I’ll be disciplined enough to keep this thread about support.

About two years ago, I felt our AI was probably excellent deep stack AI. I thought the monumental effort to make a computer play good deep stack poker (i.e., the foundation for the software discussed on this thread) was nearing completion. I wanted to know how good it was. Moreover, I needed to know where its leaks were.

I came to conclude that this was the primary problem the University of Alberta had (i.e., they could not test their AI for real money play). They had no way to know if they had leaks in their program. Imagine trying to learn how to play poker, but not playing for money. This was what caused them to fail in my opinion after a major decade long effort. I did not foresee this problem. So I had to make a decision: (a) give up my dream to make great poker AI and great training software or (b) find out how it played against people.

So rather than producing crummy training software, I decided to play online poker using our software as an advisor. To do it quickly I worked w/ some others. I told myself I was not breaching terms and conditions because other “advisors” were allowed and people were pushing buttons. I also tried to convince myself that it was okay, because I was using an analytical tool that I had created. However, I quickly learned that I was wrong. The first site closed the accounts and took my money. Then I tried it on another site and had a similar experience. I lost a lot of money on this effort but I did learn a lot.

Since this time, I have discovered a far better way to test our AI than using it as an advisor. (I wish I had thought of this before, but I didn’t) We replay hands played by great pros. Then we look at the deviations similar to the way our program works.

I do feel bad about having used it as an advisor, but I did. And what’s supposed to happen happened. Presumably the sites returned the $ to people, and no one was hurt. I have not done this since nor will I – particularly now that I am no longer a lone gun, but a part of a professional enterprise with employees, a partner and investors, and have a better alternative. To test our AI we can benchmark it against pros. This has proven much more effective in improving our AI than my early and dumb effort to get real data.

So I made a mistake, paid a price, learned from it, and have moved on.

I think what we'll do is give people the option to change the number of simulations.

However, maybe we'll automatically do more simulations if a decision is close.

For flop situations, maybe we could start with trees like:

flush card comes
flush card doesn't come

you make your straight
you don't make your straight

or maybe we just skip flop analysis because the tree's so crazy

If you know the StoxEV folks I'd love an intro.



You are right about frequencies. A prior version of the software was much more like you say -- letting cards determine our mixing up of our play much more. However, we didn't like advising customers in situations to bluff 100% or bluff 0% (thought this still exists in some situations). It seemed somewhat misleading. For example, with one hand our AI would want to bluff 100%, but with a slightly different hand it would want to bluff 0%. So we change it so that it was saying -- bluff 75% in the first case and then 25% in the other.

So part of what the advice is currently saying (or should be saying) is in better bluffing situations bluff more often and in worse bluff less often (and of course if it's a crappy time to bluff do it 0%). It might actually be very slightly lower EV -- but I think it's a better learning aid this way.

What do you think?

Was this hand that had the weird differences the J8 hand? If not, could you please PM the hand history to me.

We tested the deviations, and we get them varying rarely more than 5% of the pot. So maybe there's something unusual about this hand that has more variance. I'd like to analyze this.

So your first response that you didn't was a lie, but now you're telling the truth. No, I posted because I know people who have plenty of information about you, such as these individuals (I don't know who posted in your support forum):
http://pokerai.org/pf3/viewtopic.php?f=3&p=34490

To claim you needed to test in real money situations is B.S. I can't believe anyone would believe what you're saying. Neither would I believe anyone in the 2+2 community would support your software, or provide feedback for you if you've shown a past of botting on sites. What you did was illegal, and a huge online poker no-no.

In the past most allowed Calculatem, etc. I think some allow Magic Hold'em today (e.g., Full Tilt) which gives real time advice. While it will give you advice during play, it isn't very good. That's why I think they allow it.

However, given the quality of our advice, they would not allow it.

I'm hoping that we can find a way to make this capability work for you though.

The AI recommends stealing w/ a wider range of hands with deep stack sizes; however, you rightly point out that it shouldn't make a difference because the blind blinds stack being a bit smaller, but the small blinds stack being a bit larger -- probably nets out to zero impact in this hand.

Is there any chance you could turn off the advisory stuff and offer the rest of it for cheap like $20-$25?

ie just have different hands and situations and plays by different opponent player types as it does and you have to make a decision .. just to practice your own strategy and have more (unlimited) time to think through situations which you don't have in a real game for when you have an unusual situation that you are not sure about?

sorry if this disregards your main idea of a brilliant AI idea which took 3-4 years, but I think it would fill a need and net you some money from those people (such as myself) who would be more willing to play $20-$25 without the advisor, but not $99 if we are aready fairly strong in poker and not looking for the advisor in the first place.

Regards

for your purpose, i would recommened poker academy instead.

Poker Academy costs almost as much. So that is not a better option.
Plus what I like about Holy Grail is that villains do spazzy plays similar to what I find players do at real tables.

I think that's a reasonably good idea. We have not thought a lot about pricing, but making it so people who want one part but not another should be able to just pay for the former.

We've also thought about making a permanently free version.

Thank you for the idea.

BTW.. If you are a lot better than our AI, then here's how you may be able to beneift from the program to still improve your game (even though it is not as good as you).

(1) You can import a bunch of hands and find out how profitable your different moves are. For example, I some great pros who still learned a lot from HEM and PT3 early in their career (while they were still damn good).

We don't tell you much of what PT and HEM tell you, but our software can tell you some things you won't find in PT3 and HEM. You can find out how profitable your bluffs are in different situations. You can compare the profitability of checking the river w/ a very strong hand to value betting it (in different situations).

You can see how profitable your c bets are into 3 opponents.

You can see how profitable your 2nd barrel bluffs are and how frequently you take different actions (which is harder to simply know).

(2) After a session, you can review the 20 hands our AI thought you played poorly. While most of those hands may be mistakes of our program instead of you, you might still find a few to learn from.

Also, while you won't be able to improve your game on a % basis nearly as much as a beginner. A tiny % improvement in your game can be worth a lot if you are playing a lot and/or playing small stakes.

Marketing pitch off.

Regarding Poker Academy (for those of you that know it)....

They have some features we don't have that would be very easy to add (e.g., set up a board and find out EV if all opponents see show down). This sort of thing is very easy to add, but we are also trying to keep our software simple for beginners.

If they have some features that we should add please tell us.

regarding... letting users set up situations and analyze them....


What do you think about this idea...

We force 15 seconds of time after you set up a hand to analyze and when you can analyze it? So even if you've developed something to immediately input all the information, you still must wait 15 seconds before analyzing it.

Seems to me that should solve the problem, no?

Should we use these graphics?

They look quite good to me!

Maybe we give people the option to use what we have now or something like this.

Do you have any other suggestions re: graphics?

Let's talk vvolf69!

Could you please make it support Pokerstars/Fulltilt.fr ?