Playing with the senarios if it 3 bet from utg+1 9 handed then it the wants to fold KK to a btn cold 4 bet, but it wants to call in the same spot with 77.

I battled snowie again today for another 490 hands 6max. This time i won 425 bb, after losing a 400bb pot with a combo draw in the last hand. That's almost 100 bb/100! I played even more aggressively than yesterday, 40/39 with 7.9 AF and obviously an insane 3bet ratio. Especially as stacks get deeper, Snowie is in deep deep water against a super LAG monkey style. If i could play 6max against Snowie for money, it would be the juiciest cash game I have ever played.

Correct me if i´m wrong here, but in the longrun, if you go up against a program that plays with balanced ranges and frequencies and you show up totally unbalanced you will loose.

Its actually kinda funny that you think you can exploit it because it "bluffs too much" or is "too passiv" to your aggression, whilst all it does is it plays close to GTO. And in this regard, there is no bluffing too much or being to passiv.

Not saying it is perfect, and it certainly has some weird moves that might look questionable, but overall it looks pretty solid..

Yes, if.

It would be funny if Snowie had been proven to play a good approximation to GTO, but that hasn't been demonstrated yet. Given the hardness of that problem it almost certainly isn't playing close-to-GTO. If it isn't, it has leaks, and since it can't adapt, the best way to exploit those leaks is via totally unbalanced lines. If you find a vulnerability, like folding a bit too much to aggression, you can hammer on it.

The point of being balanced is to be unexploitable. A program can play an unexploitable game (GTO) and be completely unable to exploit you.

Most programs can't adapt well, so you don't need to be balanced. For example, snowie doesn't adapt when you 3bet him 100% of the time.

If anybody here really thinks snowie is near gto or world class or whatever you want to say feel free to set up crossbooks with me where I will play it while you watch. (HU only)
We can crossbook stakes/sidebet after a certain amount of hands/whatever you want.

I have not played hunl regularly in a very long time and would be a dog I imagine to a ton of people even at the 5/10 hu nl level nowadays.

Read this forum a lot, this post actually interested me enough to create an account.

Semi new to playing poker, have a year or so of experience.

My main expertise is Computer Science as I am approaching my masters thesis within the next half year. Over the past two years, I have learned quite a bit about AI, neuronetworking, information processing, and functionality. More so on actually creating cars that drive themselves, but also had the pleasure to research and implement my own thoughts into super computers and their processes.

Enough of the bull*****. The important parts of this entire debate are:
1. Is GTO actually possible to achieve?
The answer is a shocking 'no'. Reaching GTO is like running after infinity and catching it, you CAN'T, it's a THEORETICAL term stating something that exists on paper but will never actually happen

2. Can an AI bot ever be close to GTO?
The answer is 'yes'. Do we have the capibilities to create an AI of such that can reach this goal? 'no'
Obviously, the problem comes from the variations of gameplay that can occur. It's actually not an infinite amount of situations, rather a massive finite number of situations that can occur. I have not seen the snowie bot code, but there are a likely large set of algorithms comprised of setters, mutators, and action listeners that are consistently used to determine a plan of action for the bot. There is an issue with this though, the bot cannot adapt to each individual situation and cannot continue to learn. If the bot is unable to adapt and learn then it cannot reach what could be very close to GTO because it will simply revert back to the individual algorithms, where the bot is updated with the situation and simply reacts to the action given its current coding.

Because GTO requires taking every single aspect of the game and using formulas to create an estimated decision, it requires a bot essentially able to be able to make its own decisions, rather than given directions from the coding. We do not have this yet, or the ability to read every individual aspect of the game and take each aspect into account properly in creating an educated decision

Your statement is provably false. It is true only in the sense that the methods that are typically used to approach problems of the size/complexity of texas holdem have the property you mention - they try to converge on the solution but may never find it.

But given that the solution exists and the solution space is finite, it is guaranteed that if you look long enough, you will find the optimal solution. It may not be *practical* to do so but it is definitely *possible*. i.e. it might take a trillion years of all the computing power in the universe.

But you also might get lucky and find it much sooner. Anyway, it's not *impossible* to find it, just not that likely.

Also, I do think it's much much easier to find "nearly optimal" solutions. A typical example of a non-polynominal problem is the travelling salesman problem. With more than a handful of cities the problem is intractible to solve by brute force but there are a number of simplifying steps that can be taken to give a nearly optimal solution (to within, say, .00001% of optimal) that can be computer with no real difficulty.

But that also doesn't mean that anyone has arrived at a near optimal solution, nor that snowie in particular has.

GTO or not that HU Snowie is insanly hard to beat. never seen anything like it.

Then I guess you should take Eskaborr's action

just played snowie like a monkey, basically winning almost every hand.
I raised snowie everytime, except when he was aggressive or re-raised me. but if he called I kept betting. I did raise every frist bet on flop and preflop 3bet or raise.

pretty much win every hand and was up a buy in quite quickly.

but the problem when I did not win the hand it became quite expensive. so don't think this monkey thing works.

over long time my stack stayed between 200-300 (in 1/2) while snowies stack rose and rose

I agree with you, I was not actually thinking as scientifically as I should've when talking about this.

The only concern I have is the fact that GTO is essentially perfect play. It is impossible for something to be perfect, but in the truest sense, nothing is technically impossible so I unmindfully ruled out the possibility. Almost like the probability that I run through a wall and re-emerge on the other side. So unlikely that it's neglectable is what I was trying to get at.

No, you're still categorically wrong here. Perfect play is not impossible. There are many games for which perfect play is trivially simple. Poker is a more complex game so it's hard to "imagine" perfect play but it most definitely exists in the real sense, not in the purely theoretical sense. It's not some kind of technicality, poker belongs to a class of games with GTO solutions.

You say you've studied cs and specifically ai. You know gto is just a poker term for a Nash equilibrium strategy right? Are you saying you don't think the game of poker contains a single Nash equilibrium?

According to the Pokersnowie site there are now 7 players who have played 5k+hands that are winning vs pokernowie. They can if the wish request to be identified on the site.

If any of them are on this thread claiming big winnings vs Pokersnowie mybe they should take them up on this.


http://www.pokersnowie.com/blog/2013...d#.Ur2x87nuORs

Just to clarify, are we considering perfect play to be the best mathematical decision for the situation? Or are we considering perfect play to be the actually correct decision regardless of the mathematics? Like folding a set of kings on an AxKx5x flop when you are likely to have the best hand but your opponent has a set of aces. Perfect play would be to fold here but mathematically the best option would be to continue and get the money in. Perfect play is just not realistic but I might be blending it with optimal play on accident. I've read a few articles, I just feel like countering your opponents every single play to give them the least advantage possible is just extremely unlikely and cannot be done - except on paper

I believe poker has a Nash equilibrium, what I was trying to get at is it's not realistic to even consider a Nash equilibrium because that would state player X makes perfect decisions based on player Y's strategy, while player Y makes perfect decisions based on player X's strategy. The only way this could be done is by a basic bot playing another basic bot, both of which take into account each other's strategy and act accordingly. Just not possible for a bot to counter an actual person's strategy because an actual person can steer from the norm at any instance - but I believe it can be closely approached and approximated over a long period of time because we are playing a statistical game. All of this can only happen if the bot is able to thrive and learn in a neural networking environment

There's a terminology sticky, I recommend familiarizing yourself with it.

When a NE exists, playing "your" half of the NE is considered adopting a GTO strategy, even if your opponent is not playing the other half, and you are guaranteed to not lose money. This is what people refer to as a GTO solution.

An example for a simple game, like say RPS, is to pick each choice 1/3 of the time. This is the NE/GTO solution regardless of what your opponent does.

Yes, yes I understand I am saying that playing your half (either bot or human) of the NE is not possible (extremely unlikely) because you don't know that person's exact strategy and cannot act accordingly! Only a robot could feasibly make this remotely possible as it would take a length of time (not exactly sure how long, but a while) to make decisions outside of what is mathematically correct to make the best possible play every single time because it needs the capacity to adapt to a player's style

You don't need to know the other person's strategy, any more than you need to know the other person's strategy in RPS to come up with the NE solution. You find a NE pair, and your half of the strategy is a GTO strategy.

There is, by definition, no adaptation required. Please, read the terminology sticky, this is all quite well laid out in there.

Went ahead and did the old wikipedia look up real quick to find an example. Sorry to run around in circles but the different interpretations of the NE just confuses me.

"a group of players are in Nash equilibrium if each one is making the best decision that he or she can, taking into account the decisions of the others."

Logically, heads up, this is AsubY = YsubA
If we take one side, which is half of the NE. We take our strategy 'A' which is based off of Y's strategy. Put simply, why doesn't half of the NE consider Y's strategy

A NE is reached when neither party can individually improve their outcome by deviating their strategy unilaterally. In that sense a particular part of an NE pair is linked to the other part. But you don't construct a NE by taking a *particular* strategy and finding the best counter-strategy. That would be the maximally exploitative solution against THAT strategy but it would not be an NE, because your opponent could improve his lot by altering his strategy, and then you could improve yours by altering your strategy and so forth.

At some point, you'll reach an equilibrium - that's why it's called a nash equilibrium - where neither of you can improve your outcome by changing anything.

If you take your side of of the NE and play it against an arbitrary opponent, you can not lose money (in the long term). You could make MORE money by playing a different strategy but that's not the issue at hand.

Consider the NE for RPS - which is 1/3 R P or S each. No matter what your opponent's strategy is, you can not lose, only tie. If your opponent's strategy is, say, to always pick rock, then you'll obviously just break even. You could make more money by always playing paper, which would exploit this particular opponent, but also leave you open to exploitation.

It's not that the NE is the "best" approach to take - it isn't. Except against another GTO strategy, there is always a better strategy to play against a specific opponent. But a GTO solution is good for a few things:
* knowing how an opponent deviates from GTO tells you how to exploit it
* playing a GTO strategy ensures you can't be exploited - against an unknown opponent this has obvious advantages
* against a non-GTO opponent, you will almost certainly still make money. Don't be fooled by RPS, which is a degenerate case where the GTO strategy can not make money under any circumstances.

I hope this clears things up.

Damn I came here hoping to find some fanboys that felt for the GTO 25bb/100 advertisement to set up a prop bet but I'm not the first one :P.

I tested the soft a lot since yesterday, a bit over 5 k hands. I was surprised to see he does adapt pretty quickly to some things (change of gear in agressivity, cbet and barrels frequency etc), although still slowlier than a human can do. And he does adapt to more specific stuff, like bet frequencies on paired board (at start he ll fold a lot, but he can end up calling multiple barrels K high, check raising air or draws or even check raising K3 993 and stuff like that) but he does so very very slowly compared to a human... He does try to balance his range against what he considers your ranges to be.
But overall he calls-and-folds too much, he is too passive, he calls too many hands out of position (even against 3 bb opens) and he's pretty disastrous.
And well the simple fact that he does adapt to opponent tendencies (which is confirmed in the faq actually, in the "inconsistent ranges" answer) means that he's nowhere near GTO (if such a thing does exist in NL, I have some doubts).

About the 25bb/100.. I'm actually losing out of these 5k hands mostly because after a while during the first session I started testing his reactions to stuff like overbet shoving rivers 500bb deep, 3beting all in preflop any 2 etc. I'm pretty sure most of the players that test it do similar stuff. But if playing seriously and once knowing a bit his weaknesses, he doesn't stand a chance against most thinking players, wether they play nit or hyper lag.
His "hands analysis" ability seem quite useless as well. I haven't checked all the possiblities the software offers, there might be more useful stuff.

Pokersnowie doesn't change how it plays based on who it's playing, and the section "PokerSnowie seems to show inconsistent hand ranges – why is that?" makes no such claim.

fwiw I used the trial to import sessions and it called me world class. I've been winning at 0.5bb/100 over a few hundred thousand hands at 200nl zoom and snowie lost to people my level and higher. This makes me think it's only really beating micro players and people as above that are messing around. Pretty misleading.