Anyone else having issues with the "Challenge" feature freezing up on them this past week?

Yes, they look very different and they become a bit wider.
The calling range also increases a little bit.

Wider (?)

what do you guys think of snowie 3betting AKo in MP vs a UTG opener?

When UTG plays AT+ and all Axs, it makes perfect sense. Supreme blockers + equity vs the hands that call = win.
Against narrower UTG ranges, it makes less sense.

I know snowie changes a bit depending what stake you choose...
How close does it stick to the advice in the preflop advisor for 6-max?
When you play it of course.

I'm not 100% sure about this, but it doesn't look like the website version of the PFA has been updated since the new AI was implemented. As I recall, Snowie used to play exactly the same as the PFA if you did a 6-max 100NL challenge, but now it's pot-raising the button if you do a 100NL challenge, while the PFA still recommends a half-pot open.
The interesting thing is that the (tight) ranges it plays in the challenges are apparently the same as those that it recommends for raked games, but there is no rake in the challenges. I think this makes the challenges beatable.

To be honest. I never really even consider rake when deciding to play a hand preflop, but I guess it does make a difference.

Playing with Poker Snowie for about 4 months, mostly FR and 6-max freezouts, with short-ish (40 bb) stacks. Sometimes try 25 bb stacks as well.

I think in these games snowie is exploitable, because it plays rather passively and will pay off on your made hands too often. I am up 17bb/100 over ~11000 hands against it, though even that can still be variance, of course.

My observations:

1. Snowie plays rather tight pre, and it's not aggressive postflop. Its AF is around 2, and it frequently sees more aggressive play as a mistake.

I wish it had some settings which will allow it to play looser/tighter and more aggressive/more passive.
Unfortunately this is not feasible for a neural net bot. It has converged to a certain equilibrium, and that's all you get. I have a feeling that different initial conditions in training may have produced a different equilibrium.

2. It is a good tool to think about ranges and constructing the hand. When it finds a blunder, this is at least something to think about.

3. Its river play is quite strange, probably because by the river it came up with rather weird ranges for the opponent. I think the turn play is more reasonable, but ymmv.

4. It is more useful in hu pots postflop, than in multiway pots. It gives too much respect to pre-flop callers, it seems.

5. It is a very good warm-up and de-tilting tool.

That's very impressive but I'd like to see proof. Can't possibly take it at face value even with variance in mind.

LOL Why do you need prove. It is a freeroll toy game. LOL

There really is no way to prove it. If you post a picture there is no way to prove it hasn't been doctored. Same thing with graphs for real games. The only thing you can go by is trust.

In that case I'll have to assume he's lying because it's extremely likely.

Its normal to beat poker snowie by more than 17bb per 100. Just because you can't do it does not mean he is lying. Why would anobody brag and lie about results of play money winnings. It is more likely that you are the one lying. Sharkscope and the chart you posted awhile back do not match and sharkscope cannet be rigged.

I'm playing with 50/100 blinds, so this is 17bb/100. The filter is to select all 6-max and FR games.

snowie was trained to play 50-400bbs deep, so basically it doesn't know how to play shortstacked. and its bet sizing model is not designed for that, for example it's never correct to minraise 10bb deep - it's limp/jam/fold but snowie inherently doesn't have such options. and you play 25-40bb deep. what's your point?

@yegor
If I'm not mistaken Snowie does train with small stacks as well. It is true that Snowie plays his weakest game in full ring short-stacked, so much higher winrates are attainable, especially over a small sample.

Knowing that a 17bb winrate is attainable over 11k hands the next question is how much it will drop over a significant sample (I think it certainly will, 17bb seems unsustainable).
Lets say ubaldus' true winrate is 0bb then there's about a 1.8% chance to achieve 17bb+ over 11k hands. With a true winrate of 5bb the chance would be 7%
So there's a high chance (55:1) that ubaldus has a winrate of over 0bb against Snowie.
Of course it's still possible that his true winrate is =<0bb (only 1.8% though)

Edit: That's presuming that the screenshot is not doctored. I remain very skeptical.

@ETBrooD

I personally find it very unlikely that someone would waste the time to doctor a picture up to prove they can beat Snowie. I believe Ubaldus and I completely agree with point number 4 in his post.

I'm not accusing him but people have done this before numerous times so it's not that unlikely. It is not reasonable to assume a position without more evidence.
I'm leaning towards him actually having achieved this winrate but 1) I'm not convinced and 2) with a larger sample his winrate can absolutely drop down to under 5bb (which is still an impressive feat).

no it does not. but even if it does, whatever it may learn will be far from optimal in real world just because of its bet sizing model(ie 1/4 pot, 1/2pot,pot, 2 pots). it should be able to open-jam, jam over cbet, minraise-jam etc etc to play reasonable shortstacking strategy but there are no such lines in snowie's decision tree in the first place, so it will never be able to learn that. it's like playing plo vs someone who can use only 2 of his hole cards.

why would it be impressive though? for all we know right now, there has been no evidence presented that snowie is actually capable of solving poker or even improving its strategy over time;, there has been some anecdotal evidence of obvious flaws within the strategy; and there has been a bunch of misleading marketing from their side. it's much safer assumption that snowie is just a bad bot instead of an opponent that you should breakeven or lose against over a larger sized sample.

People do not exactly come sprinting screaming they can beat Snowie. If they did we would hear of them, that's just human nature.

But I and and a lot of posters have stated how easy it is to beat this dysfunctional toy bot but you just choose to ignore it or deny it, actually saying that pics showing profits are doctored. How silly is that, who is actually going to doctor play money results. Absurd.

They claim that Snowie played trillions of hands with a variety of stack sizes, so I assume it should be able to make some adjustments. Of course it does not understand Nash push/fold equilibrium, so it gives up some EV here, but that should be well under 5bb/100.

I would not bet I can replicate 17bb/100 in another 10000 hands, a part of it may well be due to favorable variance. But I feel fairly confident about 8-10 bb/100 under these conditions.

Why am I playing FR with short stacks? Because it's exactly the setup I am currently interested in - these days I play mostly live MTTs, and with fast structures you get under 40bb real quick.

The ideas under Poker Snowie are interesting, but they need to compete with the other bots (at the very least) to back up their claims. As I said before, their neural net found a certain equilibrium solution, but it may be a local optimum, and very far from a true GTO. I know they did an awesome job with backgammon, but it's far from clear whether their approach transfers well to NLH.

Snowie is just one profile. So the amount of practise you get is too limited.

The profile has some expert knowledge, especially preflop that you get free from their site plus more with the trial. It does split its ranges on all streets oop and ip, that is grossly correct, but in a snowie way and with a sticky deck on the turn and river. It turns its preflop calling hands into bluffs, that is not all bad. It has a couple of tricks, patterns, non balanced at least half so, postflop.

You should not be using its scenarios but you can get some ideas. They are usually not balanced on any street, but the ideas are somewhat there.

You should have and use our own expert knowledge and style, something you wont mostly learn from snowie because it mixes up so much and isnt balanced enough, and you will be thinking many things snowie doesnt. If you are not of legal age, you can buld some training after you have studied the books.

I used to play Turbo Texas Holdem, limit holdem software of the past, building all kinds of profiles, and then i could play any handed vs them and run simulations. It was an easy win, random cards, and it played well, any profile i wanted. Human always plays better, in spite of that good profiles. Snowies profiles are not any better, and actually worse, but it cheats to look better. It isnt worth to learn much from, being just a toy, for one profile training and being like one book of knowledge, that would be better as a book, playing being a waste of time if you can play real also, and even as a book, it just adds some, or the knowledge is out there if you have read the books.

But if you opt to play it, just play your own game and you will get it improved some. You will also adjust better to other players who play more or less like snowie and will read their game better, including the bad plays.

There are also gto softwares, but they offer only a technical and up to a joke balance. Again just something extra to add into your game and it can also be free, or you can go for more, but it is not going to change much the way you play because you are balanced enough and adjusting technically and non technically, as the way gto plays is just a collection of ideas and opinions in that case also, and could be one book.