Well it can in theory...it is just the game we currently play is in $ and cents and normally starting stacks are an artificially limited i.e most commonly 100BB....but we can set up a poker game where you can infinite different number of bet sizes.

Dear Oracle3001,

as you might have understood, I'm taking this criticisms very seriously and as I previously stated, we are reviewing our website in order to make sure that no misleading statements will be there anymore.

I appreciate your feedback and you might have noticed that most of it was in fact taken into consideration, among other little changes we applied. Any additional feedback from the 2+2 readers is welcome.

Now, about "not presenting facts to back this up", that's only partially true, but we appreciate that the community needs more solid proof. Be sure this is one of our top priority: we are working on several areas to convince you about the Snowie strength and you will be updated in the coming weeks and months about our progress. A special thanks to all the readers who have given us suggestions about this.

In the meanwhile you still have the possibility to build your own opinion by playing against it, evaluating your hands and situations, looking at its hand ranges and think if the suggestions make sense to you.

About the challenges, like I said in a previous post, we are setting them up in way that will avoid any doubt about the random generator being manipulated or Snowie taking advantages by supposedly know the opponent hole cards. I realize that some suspects could still come from people guessing that we are not reporting the real grouped results, and some doubts could arise also from the fact that it's a non real money environment. But it's pretty clear that every single player has the possibility to build his own opinion.

Best Regards,

Roberto Gobbo
CEO - Snowie Games Ltd.

I really don't mean to keep going to town on you and I notice you have altered some claims, but misleading claims / claims you aren't backing up are still littered across the site e.g.

"This final strategy is close to Game Theory Optimal (GTO) Poker."

http://www.pokersnowie.com/about/tec...-training.html

I think it is pretty clear from an academic who specializes in this field that given your limited bet sizing simplification, it is frankly impossible that this claim is true. Unless you are talking specifically about a semi-limit version of poker, where only these restricted bet sizes are allowed, but even then you have no facts to back this up.

If you had just stuck to the tough AI / bot claims from the beginning, I don't think you would be catching the kind of criticism you are from the 2+2 community.

It seems there are people out there who think your AI is decent and tough to play again, but again as an expert in this field has stated, a human thinking an AI agents strategy is tough to beat doesn't in any way, shape or form mean that it is close to GTO.

And if you want real "street cred", as other has said, enter the bot competitions.

First, I would like to report here a link from our blog where we speak about bet sizes:

http://www.pokersnowie.com/blog/2013...ndle-bet-sizes

As explained, Snowie uses the bet sizes half pot, full pot and two times pot. These are only three options compared to the many possible options. Nevertheless, there are good arguments why Snowie's disadvantage with this limitation is not very big.

- First of all, the three bet sizes cover the whole range, from relatively small bets to significant overbets.

- Then, in many cases the EVs of certain bet sizes over the whole range are quite similar, due to equalizing effects. If for example your EV for some hands in your range is much higher for full pot than for half pot, most likely you can bet a greater number of hands with the half pot bet size (where you gain less per hand by betting compared to checking, but more often).

The PokerSnowie Team

It might help trying to clarify the meaning of this sentence: the principle of our learning algorithm (which is explained on our web site) automatically leads to a GTO approximation. There are no human hands used for training Snowie.

Additionally, if you look at Snowie's ranges, you normally find a good balance between bluffs and value bets. If Snowie was playing an exploitative strategy, there should be either no bluffs (if the opponent calls too much) or very many bluffs (if the opponent is too tight). A balance only makes sense if the opponent is unknown or is able to adapt to your strategy.

As we wrote on our website: "PokerSnowie reached a strategy that is so robust and balanced that it becomes very hard for the agents to find exploitable leaks in PokerSnowie's play".

About being close, we already stated that we don't know precisely how close we are, but we think PokerSnowie is one of the best approximations currently, and like we already said, we will work hard in the coming weeks and months to back this up. At the same time, the training and improvement of our AI keep running.

Best Regards,

Roberto Gobbo
CEO - Snowie Games Ltd.

I have read you website and fully understand the method you claim to have used to form your AI. I have reasonable knowledge of this field, although my academic expertise lie in a different area of computer science.

You can keep twisting around like that, but doesn't make your claim true. Any talk of involving "if you play it, you will see it looks balance" is frankly a nonsense argument when defending a claim of "close to GTO", that is if you want to be taken seriously on an academic level.

You can try and weasel what does "close" really mean, but if you really have the attitude you claim you have, you still have a misleading statements on your website.

Compare and contrast the way in which you defend your product vs the academic expert who chipped in.

At the end of the day, its your product and you market however you see fit. But don't be surprised when people point out some of your claims are marketing BS and don't stand up to much.

You may take this as some massive negative rant against PokerSnowie. It isn't, it is purely pointing things out in relation to your marketing.

You may well have the best poker AI. It is certainly a professional looking product, plus website and you guys are clearly well funded to support the sort of advertising you have been undertaking.

My humble advice would be stick to talking about / proving that, especially to the knowledgeable 2+2 crowd.

I'm actually curious even how close to GTO is if you allow the bet size limitation - that should actually not be too hard to compute using the information provided above by FullyCompletely.

Beyond that, I am curious about the expected difference between a strategy that is GTO given those constraints and one that is GTO with unlimited bet sizing.

Without a demonstration of the first, I have no interest in Pokersnowie's GTO claims - and I feel it's something that COULD be easily demonstrated. I am not interested in how well Snowie does against human players, marginally interested in how it does against the best bots, and very interested in how it scores vs GTO.

I understand how people don't like the GTO claims snowie is making, since it cannot ever be actual perfect GTO but only an approximation.

However snowie is not a science project. Snowie is a product. I am a customer.

For me I am looking at a way to learn what GTO is. As I am human, I can never learn GTO since my brain is too limited ( I will never use unlimited bet sizes).

Snowie is the first time, I actually have a way to see GTO. Now obviously the snowie GTO is not the actual GTO, however they seem to have found a way to:

a) find an approximation of GTO
b) build a product that lets me see their model
c) compare my game against their model and use for game review
d) build a trainer

In other words snowie has found a way to turn a science thing (GTO) into something usable for poker players. Of the above a) is actually not quite as critical as people think. As long as snowie beats like 90% of winning players that is sufficient for 99% of players (which is good enough to be an awesome product for the snowie team).

I have seen many bugs and mistakes in snowie and that is troubling. But at this time arguing about the GTOness of snowie is like aguing that the first car was a stupid invention because it needed gasoline and was slower than a horse (yes this really happened).

However if what snowie is doing somewhat works than this is revolutionary to how you can learn and understand poker.

Can we please also use this thread to help understand snowie better, how it works and how i can use it for my poker game.

knircky,

I don't think that over the entire course of this thread, you've ever really understood the objections. You keep saying things that are on their face quite stupid.

You have no reason to believe that snowie is even a reasonable approximation of GTO. Yet you keep speaking of it as if this is the literal truth.

It would actually be possible for them to quantify how close it is to GTO. They decline to do so. They decline to even address the issue.

The fact that you continue to state that it's an approximation of GTO is pretty staggering. If they're not paying you, they should.

Love,
Rusty

They're also obviously just ignoring questions about testing the AI by competing in the ACPC.

Combined with the fact that their advertising materials have essentially been bold-faced lies, I don't see any reason to give them the benefit of the doubt here.

+1.

Please provide proof, or at least explain the principle of your learning algorithm (the explanation on your website is not sufficient to make this claim, nor have you provided a benchmark of any kind).

Additionally, as in additional clarification of the claim on your website?

You feel snowie claims they have solved GTO without actually proving that they have.

Because they use this for marketing you feel they are a scam since they claim something that cannot be verified and cannot be true.

Am I missing something?

Thx, great and useful information.

I understand for most 2+2er are here to have a discussion and win the arguments. Insulting people is very useful for that.

I am here to have a dialog and don't care about discussion.

I know that snowie cannot play GTO. Its written on the snowie website that it can't. I actually red the website.

I do assume that snowie is the closest to GTO I can get and use it as such.

I find telling snowie that their product is not precise GTO not so useful (since that is a given).

Why are you assuming this?

Why not? Do u know anything better than snowie?

Here are some things that lead me to believe snowie is pretty good:

1) Their product is designed to be close to GTO. Their approach makes sense.

2) Their learning gto vs calculating it seems to be the most efficient way to get close to GTO. Their cloud approach as well.

3) they make similar abbreviation that I would make (i.e. 3 bet sizes, though i also use 2/3 pots)

4) i think i was able to learn HU with snowie pretty well

5) i have my game drastically due to snowie and i think it has shown in decent results

6) i have not seen anyone actually show that snowie is not gto or that their strategy is exploitable.


If snowie is somewhat close to GTO than this would be huge. Though at this point i dont trust it too much, i.e. try to evaluate situation myself but try to understand why snowie might have a different line than i would consider standard.

I would actually appreciate if people could show where snowie is wrong and why. But so far people tend to diss snowie vs. providing information that actually would be helpful or show mistakes snowie is making.

Yes. You keep claiming they are "near" GTO or a "GTO approximation". You have no reason to believe this. It's tantamount to thinking acupuncture can treat cancer.

Dialog(ue) and discussion are basically the same thing, so, uh?

Why? Try to answer without saying "because they say so"

I find it interesting that they have a rep in this thread who's not interested in telling us how good it is. You know, like a measurable number. Something it should be possible for them to do.

again know any better GTO model? Or maybe u can show where and why snowie is exploitable.

about as close as GTO and exploitative play.

What better way is there than a neuro network to solve this?

Maybe u can explain why this is a bad approach. Again looking for a dialog type response from you.


I would love to know this too. Don't see how it can be solved though.

I hope they will compete with some other bots or humans.

If they are just a tad exploitable it should be easy to beat them since at any point you know exactly what snowies range is and how snowie will react to ur action.

Like any human should be able to beat snowie since we would have perfect information against it. Snowie team has said this btw also.

Looks like plenty of ignore lists ITT aren't up-to-date.

Dear pokersnowie,

While browsing your website, I found this:
So I decided to take the challenge, but after 2341 hands and 28BIs won, it became quite boring to fight a losing battle against such an impressive AI. However, I only have 2659 hands left to complete the challenge, so let's assume I'm going to fold every blind.
Being up 2823.72$ over 2341 hands, this means I'm going to fold 2659 hands, which should cost me 1330 + 664 = 1994.5$ (actually, less than that as snowie will fold some small blind, and I may open shove AA).
This means that by folding the next 2659 hands I'm still up 829$ with a winrate of 16.58bb/100.
So, can I still be acknowledged as a Challenge PokerSnowie winner ? (please ! please !)

Obviously, my current winrate is well above 100bb/100. This might lead some sceptics to believe that pokersnowie may not be a good approximation of GTO. I guess there's no point in discussing with these narrow-minded idiots.

However, what worries me is that pokersnowie ranks me as a "Beginner" and gave me an error rate of 72 (maximum error rate). In other word, snowie accuses me of clicking buttons. Ok, that might not be completely false, my Aggresion Factor of 40+ may not be optimal, but it's not my fault if I like to bet into weaklings !
Snowie also said that I gave up 4 573.42$ in EV. I personally think snowie gave up much more than me.

I think my username was "Babar".

NB: I suck at poker, especially HU.

Most bots who enter the annual computer poker competition have proven much more than snowie ever has. IIRC, most of them (if not all) also use a GTO approach in HU.

Is their model public. it sure would be interesting to compare.

The competition is open to anyone. If snowie wants to prove the strength of their program, they have a great opportunity here.

Dear RustyBrooks,

as I already said, we are working hard to set up a framework of information which will show some more concrete proof about the strength of Snowie and its relation to GTO.

It’s not correct to say that I’m not interested, neither I’m ignoring the questions here. We just need time and we ask for some patience. And once again we thank all the readers like you who are giving us input and suggestions about how to reach this objective.

Best Regards,

Roberto Gobbo
CEO - Snowie Games Ltd.

I saw your previous post, about the random numbers and predetermined cards etc etc so that people could verify snowie wasn't cheating. This isn't what I'm talking about. IMO this doesn't even really seem to be an issue since many people in this thread have thrashed snowie quite badly (by playing bad poker, mostly)

Did you read the post by the researcher up thread a few days ago? About mechanical means to find a lower limit on the exploitability of snowie? Do that.

Or, barring that, enter the competition also mentioned up thread.

These are real, substantial things you could do to prove your results are serious or interesting.

Dear RustyBrooks,

you are right, recently we wrote a post addressing some concerns about Snowie supposedly cheating during the challenges, but this was not the post I was referring to. And by the way, there has been many posts on this topic, on other forums as well. Therefore we have worked to address these issues and from the next release, there won’t be any more doubt about Snowie manipulating the challenges.

Going back to your post and to the more relevant issue related to GTO, as I stated already more than once, let me quote: "Be sure this is one of our top priority: we are working on several areas to convince you about the Snowie strength and you will be updated in the coming weeks and months about our progress."

Therefore, once again, we are taking very seriously all the concerns about Snowie vs. GTO, bot competition idea, scientific articles and approaches, measuring the distance between GTO and Pokersnowie gameplay, and in this sense the feedback that the 2+2 readers have given us are very useful. In other words RustyBrooks, point taken, your concerns are serious and we understand them, now let us some time.

Finally, about people having thrashed Snowie, I have not seen any evidence of people having played a substantial number of hands (50K+ hands or more), who had the results you are talking about. If you believe Snowie is that bad, we are very happy to take those hand histories with the permission of the players you are referring to, and review them.

Best Regards,

Roberto Gobbo
CEO - Snowie Games Ltd.

Who would play 50k hands of play money ? I played 2.5k hands and it was already quite boring.
If you only take into account 50k+ hands samples, then you don't have any evidence that snowie can beat anyone either.

It's the second time I clickraise snowie and beat him over 100+bb/100. I may be running hot this time as well, but I wouldn't expect to beat any human like this, even over a smallish 2.5k hands sample.

It seems that the most honest thing to say at the moment is that noone has the slightest idea of where snowie stands at.