I would like to know, what a bot like pokersnowie is doing wrong in terms of GTO - is it just the fact that it is not powerful enough, or does it actually make some serious or minor mistakes in terms of current understanding of GTO?

I have been playing against the bot 6 handed (so 5 pokersnowie bots) for a few months have been imitating its play in a way, but I fear this will poison my instinctual decisions - if it is indeed a useless bot? (if it is can you explain, and is the way the bot tells me to play in 6 handed games extremely far from adequate GTO)

My goal is to be seemingly unexploitable while I sit and grind and after gaining enough information on players I will then employ exploitative techniques.

Check out the Snowie containment thread:

http://forumserver.twoplustwo.com/sh....php?t=1384182

Snowie is probably a lot closer to GTO than most 100NL online players. Whether that means that someone who plays just like Snowie would crush 100NL is a different matter entirely. (As I understand it, the Montenegro crew that were early beta testers of the software grind 200NLz and 500NLz with low winrates). This is one of poker's problems in 2015. Your opponents don't have to play anywhere close to "perfectly" before it becomes almost impossible to beat the rake. Regs aren't playing anywhere close to GTO, but they play close enough that a "GTO approxabot" can't crush them. With such low winrates, variance has a huge effect on results, so it's pretty hard to even find flaws in a strategy, unless you have gargantuan sample sizes of your own.

Snowie's "problems" that prevent it ever achieving true optimal play include that it is restricted to picking from from three bet-sizes, and that it doesn't have a large enough sample size for it to have accurately calculated the optimal play or EV in every possible situation. (Multiplayer games with a no limit structure are effectively infinite in scope). It still has weaknesses in multiway pots, for example, and it seems "weird" on the river on some boards in some situations. Pre-flop and on the flop, I think it's pretty damn solid, although its blind vs blind play remains somewhat unconvincing. (Not that I am in possession of the GTO solution to poker. Snowie just does some things that contradict itself!)
To beat current games in the 25NL-200NL band, I think using Snowie for guidance is helpful, but deviating from its strategy in order to exploit player pool tendencies (or individual player imbalances) will likely lead to higher winrates. If you play lower than 25NL, a much more unbalanced/exploitative strategy would almost certainly be more profitable.

P.S. To answer the thread title with a quick example, Snowie will bluff-catch very light on the river in what appears to be a break-even GTO manner... but in a spot where a 50NL reg would almost never be bluffing. It's not that Snowie is "wrong". It's more that humans don't play like dispassionate robots, so if you played like one yourself, you'd be making "correct GTO plays", but losing money against non-GTO players.

Thank you for this answer it is very helpful. Furthermore how would you tailor what you learn from pokersnowies bet sizing for use in live games as live people use 'weird' bet sizes so much more than online?

I'm not well qualified to answer that, as I rarely play live, but I certainly found some odd recommendations when I looked at some microstakes hands that involved limping or 5x pre-flop raises. If you're playing 1/2 live, Snowie is unlikely to be very useful to you, as Snowie mostly trained itself to play heads up pots with "standard" online bet-sizing.
Facing oversized opens (like 6bb), Snowie would play much tighter than it would against a 3bb open, because it maps the larger size to a stronger range and it doesn't expect to recoup the cost of calling if the opener is competent. In real live games, you could often call with a wide range, because you can expect other players to overcall, or because the opener is bad enough to pay you off every time you make a better hand.

Additionally, if you're playing live with lots of multiway pots and you input them as scenarios in Snowie, it would probably give you some very surprising suggestions. The reasons are complex, but it basically boils down to this idea: Humans play loosely in multiway pots, whilst Snowie is generally very cautious. Snowie would sometimes fold a hand as strong as an overpair (or even a set) in a 3-way pot, because it doesn't expect the other two players to put much money in the pot with weaker hands. In the real world, bad players routinely bet and call with very weak hands in multiway pots. In short, Snowie would make "GTO folds", that would be less than optimal against players making non-GTO bets and calls.
To put it another way, you can play much looser against bad players than you can against good players. Since Snowie trained against "good" players, it would teach you to make what look like nitty folds, that might be incorrect in your live games.
That's not to say that Snowie won't be useful for you at all. You can just revise your "Snowie sizes" up or down when you play live to account for the different game conditions. i.e. You can learn an approximation of what GTO play might look like from Snowie, but then deviate it from it to maximally exploit your regular opponents.

So could you just scale your bets up in a live game, would that still be the same EV that snowie credits it to be or would it change a bit?

And I dont understand how as players play a non-gto game it would put someone playing GTO at a disadvantage rather than give them more money, as I thought; any play that steers a way from GTO is coincidentally a mistake and thus giving the only player playing GTO or close to it positive EV even in multipots or maybe even ESPECIALLY in multiway pots.

I thought this would be true unless there was collusion or does the larger amount of players in the pot/larger amount of players at the table increase the chance of 'accidental' collusion where several players' actions collude without knowing it? or is it more complicated?

Nope. Your opponents will be playing much differently also, which will benefit some hand types over others. For example, hands that do well in mass multiway pots are much more profitable vs. low level live opponents than vs. snowie opponents, because there will be many more mass multiway pots. Also, hands that would fold to a 3b preflop are less penalized vs. a pool of human players that mostly only 3bets KK+. These differences are also why using a raise size that's much larger than optimal vs. strong players can be more profitable.

It doesn't put you at a disadvantage vs. those players. You will do better than them either way. You will make much, much more money though by exploiting your opponents' play as much as possible. I.e. you can still beat the game (prerake) if you play the same vs. crusty old nits' 3bets as you would against snowie, but you'll make more if you adjust to their very unbalanced range by folding almost everything in that spot.

There is a kind of "accidental collusion" that occurs in multiway pots. It's not well understood by anyone, but the theory is that whoever makes the first bet in a multiway pot is often donating free EV to everyone who acts after him. e.g. If it was 3-way and there was a bet and a raise and you acted third, you have a huge "information advantage" because you can fold some marginal holdings. In effect, the player that raised gives you an incentive to fold, when you might have called and given the initial bettor some value. So the initial bettor lost a potential customer (you) because of another player's action. You weren't actively colluding with the raiser, but his action stopped you from giving some money to the initial bettor. (The middle player "accidentally" saved you money, which means his action increased your EV, and reduced that of the first player).
Similarly, if you had a weak draw and there was a bet and call, you now have great odds to call and close the action, so you benefited from the player in the middle. (You could also squeeze, which makes it even harder for the initial bettor to make money).

In splashy live games, players bet and stick around with weak hands in multiway pots, whereas Snowie is much nittier. It tends to only build big pots multiway if it has a very strong hand or a draw to the nuts. It therefore will often fold top pair/overpair if two other players are putting money in the pot. In live games, though, bad players are often putting money in multiway pots with much worse hands than top pair, because they don't realise that their actions actually benefit whoever acts after them. It's all pretty complicated. I suppose you just have to get a feel for the wider ranges in live m-way play by getting experience.

http://blog.gtorangebuilder.com/2014...p-what-it.html

snowie isn't converging to GTO (otherwise they would publish it); read the big thread.