They say it played trillions of hands against itself...u r speaking about spots not played optimally like when u put in analysis of a hand u played online.You will see insane moves if u do that.

tbh i dont understand what you mean...

Quote from pokersnowie website

It is what I said...It plays against it's own ranges(neural network solutions over trillions of trials) NOT against opponent's range.Snowbot started to learn the game randomly.By countering strategies it means his own moves bot vs bot.When u sit in a training mode in a 6 max u sit against 5 Snowie bots.So u play against 5 of the best players in the world.If u leave 5 bots will play against each other and eventually they will reach a point when they can't exploit each other.If there was not a rake factor probably Snowie could solve the game somewhere in time in a zero sum environment.

so here you are referring to the scenario builder?

No ,when u import hands from HM2 or PT4 database. I say that the best way to learn with Snowie is to use scenarios and training mode.And it is easier if you have full subscription. Otherwise u will be in the dark.

honestly, i have never imported hands to snowie so i dont know how that is... i always use scenario or training. but that part

applies to the scenario builder.

I think it's about training mode but never mind.When i learn through scenarios i learn without particular combos choosing. I pick only board cards and analyse Snowie moves range vs range.

no it does apply to scenario builder. thats mostly what i use

No, you're not missing anything. Snowie has lots of quirks like that, and I think it's because of the way that the artificial neural networks learned the game. Snowie sometimes produces "weird" results, particularly in multiway pots where it hasn't 'experienced' the spot enough times to get anywhere close to a true equilibrium.

As Zuko pointed out, it will sometimes say to call with XX hand (because Snowie thinks it at least breaks even), even though it would lose 100% of the time vs Snowie's own (unbalanced) strategy. The opposite is also true. I used to play pots with it where it would fold 99.4% of its range, even though it would bluff in the same situation if the roles were reversed.

A bit like humans, Snowie's much better at playing on the early streets, due to having more experience of those.

The trick is in the hand strength button...i have checked and 99 there is considered weak with 0.52 showdown strenght which means it has 26% equity against opponent's range(Snowie formula is dividing 0.52 by 2 to get equity number).Betting strength is different because is matched against opponent's non folding range.

Thanks. pretty interesting how this thing works. Seems relatively easy to figure out what is going on preflop. It's much more difficult to figure out what the hell snowie is thinking on later streets.

i wonder if it over defends in certain spots to counter balance the spots it is under defending.

Hi, has anyone used the freeze out training option in pokersnowie? How good are the suggestions made by it? As I understand, it's restricted because it only chooses one size so it can't properly evaluate a strategy that plays push/fold with some hands, and min raises with others.

I found out that Snowie is folding 55% against 3x open hu. Isn't this exploitable? I mean, if we are defending less than half the time, then villain should raise with any two cards simply because he is making money immediately.

What do you think?

Did anyone say that snowie isn't exploitable?

It has been said that it plays near optimal or near perfect strategy. Furthermore, preflop is the most simplest street in poker to solve, and therefore Snowie should be accurate atleast on that street, or otherwise it is not delivering much to it's users.

Snowie was never very accurate. If you play like it, you probably won't make money even in micros.
It's hard to prove that something isn't a "near perfect strategy", that's why the company can get away with such claims.

I don't know why you think that preflop is the simplest street to solve though. It's the hardest one in my opinion.

Because preflop can be solved more easily by running billions of simulations. It is the street that is played most often in poker, that is why it is the most easy one to solve with this "trial and error" -approach. On the other hand, turn is played little less often and there is more ranges and betsizes to be taken into account, so it is more complicated to solve compared to preflop.

Even high stakes players have leaks in their strategies but you can be pretty sure that their strategy on preflop play is the one that is most closest to optimal.

It's non sense to say that Snowie can't beat anyone.I bet noone human can beat Snowie HU 500 k hands with same Libratus pre game rules.

It depends on how the streets are solved. If they are solved mathematically, the river is the simplest. If it's by running billions of simulations (like snowie), I guess the preflop should be the simplest one, so you're probably right.

I don't really play HU, but since snowie is folding 50%+ vs a 3x open, I don't think it should be too hard to beat it.

If it's making such big mistakes on the simplest street, one can assume that it isn't really playing much better postflop.

What r Snowie's frequencies it depends on levels u r playing.It considers rake.

If BB is folding 50%+ vs a 3x open, SB should always be opening 100% of hands (no matter the rake).
I doubt that's the case for snowie though. I don't use it anymore, but you can check.

I've used it in 6-max and can tell you that a lot of its plays don't make much sense and some of them are absolutely terrible.

To nickname disident:

This is true. However, even with rake considerations taken into account, I think we should not fold 55% preflop against 3x.

I don't know as I've never studied HU games in depth. What frequencies are the best heads up players in the world currently using?
Last I heard, the 2.5x open had become standard, and there was also a vogue for limping, but the game is always changing, so I don't know how often people defend vs a 3x open. My gut feeling is that it's probably impossible to profitably defend more than 50% OOP vs a tough opponent at low stakes (with high rake), but I'm really not sure.
In the past it was said that Snowie was stronger heads up in comparison to 6-max, and hardly anyone could beat it HU. (Even a primitive version of the AI beat Jungleman - over a tiny sample size - when he was arguably the best in the world).

Regs usually fold around 30-35% vs a 3x open.