Agreed, i'm all for discussing Snowie and trying to learn, however making snowie stronger is awful for poker.

In a 100bb 6max game, when the big blind 5bets the cutoff for 45bb, Snowie recommends to always flat with AKo and to fold to a shove on almost any flop without an A or K or trips (it tells you to fold on a 232r flop even though you get 3:1 on the call but to call on a 222 board). If checked to, it tells you to ship pretty much any board but to check behind on a paired board that doesn't have an A.
I also noticed huge range differences given small changes on the flop in the above scenario. If villain ships 55bb on a 37Tr board we should call because villain only has a pair 58% of the time but if the board is 47Tr we should fold because villain has a pair
85% of the time even though neither 3 nor 4 helps his or our range.
Can anyone explain to me the logic behind this and how come cards that don't affect either ranges have such a big effect on both players' decisions?

I've not played Snowie yet but I'm keen to find out just how easy it will be to exploit in comparison to Neobot. Could anyone make a comparison? The NL bots are not GTO by a long way in HU and 6m bots are terrible. I'm not a FL player but I couldn't beat the bot out of a 50bb starting stack. I found the bot was playing really well. I got ahead a few times but could never finish the bot off! I only play live these days but it was fun to play Neobot. Hope Snowie offers a tougher challenge at NL HU.

BB's 5-betting range in this spot is JJ+ A7s+ A4s-A5s and Snowie wants me to fold AKo after investing 22.25... Is this plain wrong or is there something I don't understand?
Since villain only calls with JJ+ AKs I pick up the pot 49% of the time and I have 36% equity the times I get called, I'm risking $77.75 for a 49% chance of picking up a $67.75 pot ~18% chance of winning $122.75 and ~32% chance of losing my $77.75 for a total EV of over +$30 compared to folding. This is a pretty huge mistake, especially for a preflop situation that it should have seen a lot over trillions of hands. How can PokerCoach get this so wrong?

there is one simple explanation

http://www.youtube.com/watch?v=qKgfiVG9ByE

Are these your ranges or snowie ranges?

Snowie's ranges.

I think snowie does not calculate pot odds. It cannot do any math if I understand correctly, maybe that's how it gets screwed up here.

Any solution reasonably close to GTO is going to have different actions depending on pot size. Not taking pot size into account is ridiculously ****ty, if true. It would mean it takes the same action if I bet $1 into a $1000 pot as if I bet $10,000 into a $1000 pot. That's obviously super exploitable and therefore can not be GTO.

It's actions are effected by pot and bet size. Not because it calculates them in the normal way, but based in the results from it's NN training.

Then it's basically implicitly calculating them and what knircky said, while perhaps pedantically true, should not affect snowie's actions. Whether it calculates them directly or changes it's actions indirectly based on pot/bet size, it's the same result - different actions for different bet sizes.

I think it just rounds up your bet size to "minbet" "half psb" "psb" or "all-in". It is obviously super exploitable.

Anyway, I reiterate my question: has anyone decent already lost to pokersnowie ?
If not, there's just no debate and we can close this topic and put snowie in the scam category.

Define "decent"?

Having read 50% of this thread and have seen no body mentioned this.

I don't know how many people have bought this app and how much they charge, I assume the company is making a profit selling this app. But wouldn't it be more profitable for the inventors of Snowie to crush every online HU rooms there are to maximise the edge it claims to have right now? If it is indeed GTO style game then it is highly unlikely to be exploitable and therefore profitable in the long term.

If selling this app is more profitable than actually gambling as Snowie, surely it is not good enough for you or me - Why would Coca Cola ever share their recipe with you?

There's sort of 2 answers to your question.

The first is, your metaphor sucks. It's against the rules of every major card room to run bots, and if they find you out they'll seize your money, ban you, etc. Wheras selling software like this is a totally respectable business. If coke couldn't sell their product, but COULD make money selling the recipe, they'd sell the recipe.

The other is, truthfully, if they want, they can do both. People who are worried about Snowie killing poker are probably wrong. Explosions in published books, coaching, videos, etc have not really made poker tougher. Poker HAS gotten tougher but it's almost always attributable to diminishing player pool, *specifically* the kind of diminishment that scoops the cream of fish off the top, leaving all the regs.

I like your second answer better because cardroom cannot differentiate a bot / human following what a bot says every move as long as you maintain being a human - Engage in chat, make a few -EV river calls here and there.

If that was true I doubt SNGs and mid-late stage MTTs would be beatable.

It's hard to prove whether there are bots that don't get caught, but there are lots of bots that DO get caught.

If not engaging in chat when playing online makes you a bot, then I must be a bot. I normally play 7 table of Zoom and never even have chat turned on.

Let's have a very broad definition of decent, I don't care. Any nl25 winner.
I may not have searched well enough, but I haven't been able to find any proof that pokersnowie could beat anyone.

I was merely stating the obvious variables, you can add bet sizes, response time, artifice mis-click, occasion sit out etc. All I am saying is that why bother selling an app when you can make tonnes of money for yourself using the app.

is there anyone who won against HU snowie over larger sample? i find it so frustrating!

No thats not the way snwoie works.

If i do understand correctly the way snowie makes the decision to do one thing or another is simply via learning vs calculation.

So it might take a look at a situation and play it several times and then simply chose the path that is most likely to be successful.

So it will differentiate between the 1 or 10k bet you have as example, but not via calculating the EV but by learning how to react best to each spot.

The end result should be the same if done correctly.

But snowie doesn't train vs an arbitrary amount of betsizes. Imo only vs a few distinct sizes. About for or so. Every other betsize it faces is approximated to one of them.

sent from phone

No matter how it supposedly calculate, I'm surprised it made a massive mistake like this. The shove was for 0.61 pot so whether it rounds to half pot or full pot it should still tell me to shove. A shove there is only wrong if we are much deeper and even then there are better options than folding against that range. I might understand if this was a rare spot that almost never occurs but this was preflop. Hope they use all the HH they get for free from users to improve the AI a bit.