I've been playing around with Pokersnowie and wanted to give my first impressions after looking through some hands with it and having heard people saying its a tool that will teach you everything and the end of poker (obviously some nl hold'em specialists there). I'm hoping to inspire some more level headed discussion of its strengths and weaknesses. It is a fascinating tool for seeing game theory and the power of computers applied to nl hold'em. It does strictly follow game theory in that there are lots of spots it considers break even between two options, and if it thinks one line is even .01 ev higher than the other it will always take the higher ev line. It certainly can be a powerful learning tool to analyze one's own game and the game as a whole.
As for the restrictions on it I think they have varying effects. The bet size restrictions of .5, 1, 2 times pot or all-in are certainly real limitations, but probably not a huge detriment. That said, it would be interesting if in the future they re-release with a 3/4 bet option. Could see whether one of our favorite bet sizes has real reason to be the favorite, or its just another options and we should be varying our bet sizes more. The one annoying thing is that I believe Pokersnowie linearly interpolates the ev between the sizes if a human chooses a different one, so it will almost always interpret your sizing as inferior ev. The inability to use two different sizes on one street actually seems like the worst limitation, though a limitation a lot of human players have in their game for simplicity reasons too.
However, I think one has to be very careful applying it to real play, as the level of precision a computer can pull off makes it play very differently sometimes from human players. Have seen many cases where a call and raise are considered equivalent, but raise is recommended 97% of the time. There is a good chance in real life whatever is causing raise to be the normal play makes it the clearly more profitable play and calling inferior, and just in general it gives you no advice on which way to go when in real life there is rarely if ever a truly neutral ev decision. I also would expect because they are near GTO play, blockers play a much higher role than in human play since knowing your actual two cards allows you to shift the weights of hands in the opponents hand lets you make very marginal decisions based on that shift that are too fine-grained for humans.
If you look at the Preflop Advisor it has lots of unusual raises going on. It recommends playing a lot more trouble hands (ATo,KJo utg) from early position. I'm thinking that part of it is the 2.25 bb raises pre-flop which would mean it should raise more often. However, I think the biggest issue is that being a incredibly precise computer it is perfectly comfortable throwing in donk bets and has fine-tuned its check-call game, two elements (especially the later) that are still some of the most difficult spots for humans to get right. Because of this it is not as afraid of playing weak hands OOP. You also may find those weak trouble hands are very marginally profitable such that any mistakes with them (something a human is bound to do, especially oop) would turn them into a losing hand. It also seems to like AXs a lot more raising it from any position and low suited connectors a lot less (maybe a combination of less implied odds from a computer, negative blockers, and because everybody is playing a lot of AXs you wind up being over-flushed more often).
I think there is a lot to learn from studying Pokersnowie, but at the same time trying to follow everything it does without adapting it to your human limitations and the limitations of your opponents could be disastrous. Simply put, playing poker in the walking a razor's edge world of Pokersnowie vs Pokersnowie and playing with humans who don't have perfect RNGs, don't have perfect recall of all situations to make same decision every time, make mistakes both systematic and random, and adapt to exploit other players is a very different game.