Durrr:

I'm in touch with Eric Jackson, the creator of SlumBot, and he says he's willing to let us use his bot. 100k hands sounds good. 2k/4k?

It's probably easiest not to play this over an existing, licensed poker client, just because it would require some programming work to make the bot interface with the software, but if that's a sticking point for you we can talk about it. University of Alberta has an interface for bot vs human play set up already. We could of course have it audited by whoever you want to make sure it's fair, secure, etc.

I think a prenegotiated buyout clause makes sense. If we agree to 100k hands, either side should have the option to buy out for 1000BB or something rather than continue at any point.

There are lots of details to work out but let's start with this.

I'm not sure how much expected winrates are or anything, but a $4 mil buyout seems a bit much, right?

You wouldn't have to add any interface....just have a human manually enter the cards/action and let the bot spit out a result. That wouldn't be necessary if he trusts you and the programers won't cheat.

In terms of eliminating any possible cheating, dealing it live with somebody durrr trusts entering to the bot audited by someone you trust would be the most secure. But it would be slow as hell and doesn't guarantee the bot isn't cheating....Kasparov famously accused Deep Blue of cheating playing under basically those conditions. Granted, the type of cheating he was accusing is prob not possible in poker.

Wouldn't that be kinda slow? I can write software that scrapes any poker client, i.e it's completely trivial.

No, that depends on the Game. GTO is the equilibrium strategy that maximizes EV. In zero-sum games it will average out to 0EV against another GTO strategy. There is a difference here - e.g. rake, blinds, position for a small number of odd hands. But the main difference relates to the other guy's strategy.

So you have supposedly studied Mathematics, call yourself the world's leading poker theorist, but haven't read and understood Nash's paper? Or you accept his "logical proof", but still don't believe the statement?

Or do you use "supposedly" in some non-standard way?

i really doubt stars/tilt would allow it anyway.

I'm aware of the problem of collusion. What is your definition of a Nash equilibrium? Nash showed that equilibria exist, no matter how many players there are. So I don't get your distinction between Eq. and Nash Eq.

I tried to point out that in my opinion we could lose even without implicit collusion, in other words, if only one opponent deviates from the equilibrium.

You can'z do anything to "beat" GTO, your opponent knows that. There's nothing to exploit.

epic ****ing thread guys
thank you all

Yeah, if they want to play it on a major poker site that both parties trust it would be trivial to code the interface (I've done it out of prue curiosity and it wasn't very hard). Of course the sites might not allow it anyway.

My concern was if they move to some other black box interface it could be hard to prove hero's hole cards or the flop/turn/river aren't known by the bot in advance. It would require some level of trust....which is fine as long as it exists.

All of this reminds me of a (quite lame) short story i wrote for a philosophy class in college about the birth of AI involving a program designed to deal and play HU poker hands against a human oppoent. It is coded to win as much as it can by playing adaptively based on what the human does. It realizes by itself that just looking at the hole/community cards is the easiest way to win (despite it's code being written based on not having that knowledge) It becomes intelligent by discovering the perhaps uniquely intelligent vice/virtue of cheating. My story sucked, but I always thought Clarke or Assimov could write a pretty good story with that premise, if it doesn't exist already.

i wish i was smart

And if it went that far maybe a couple of hours later it might fully control every particle in the universe for some unimaginable end. Good thing current hu bots don't seem to creatively seek goals.

I think this statement is unlikely to be true. It may be true that GTO bots beat terrible players with a lower winrate than good players.

lol durrrr is gonna CRUSH slumbot.
..

..

..

hey ike, can i have a piece of your action against durrrrr? or have all the action already been locked?

That must be true. Just a non idiot human player will do much better than a GTO bot against a donk who shoves every hand at the first opportunity.

True, but for less terrible opponents it's not so clear. The player you describe would be ~-730 BB/100 at 100 bb stacks vs a perfectly exploitative strategy. I'm not sure that a player who is pre-rake -20 bb/100 at HU NL or HU PLO against the field would necessarily fare better against a GTO bot than a good player. Likely it would depend on how diffuse his errors were. A human may be better at taking advantage of gross errors while a GTO bot is much better at punishing fine errors.

Demonstrably not true. See: Bots at iPoker.

Never read his paper and see no need to. I'm sure Nash proved what he attempted to prove. But since I am only 99.9% sure this means that a non adjusting unbeatable head up poker strategy always exists (based on a synopsis and the statements of peple on this site) I used the word "supposedly". I realize that word normally conveys significant doubt but in this case I definitely used it in a non standard way.

What's that?

No....there are far easier questions about chess that are still unsolved than a constructive solutions. It is not even know whether optimal play from both sides results in a win for white or a draw. Even a win for black is not rigorously ruled out....but would be a hugely shocking result.

Where on Rybka's wiki page does it even say anything about chess being solved? You realize that solving the game is not the same as beating the best human players? AFAIK Rybka isn't even the strongest program anymore.

Yeah, I agree with this. There surely exists non realistic theoretical bad players who can be beaten for more than GTO by even mediocre humans, but it is not clear if a standard bad human player will be better off against GTO or a good human.

good read guys interesting thread.


1000bb is 10 buyins which is 40K buyout