[tldr; snowie = scam (at best) - don't buy it, don't help them]
Hello to all,
I know I don't post much in here (like almost never) but I came across a situation I think I have to share to a wider audience, therefore this post.
I'm a french player, and part of "poker academie" website community.
A while back, some video there started to present and try to demonstrate the benefits of Poker Snowie.
Of course some people started to show some interest, while others were kindof skeptical.
After all this, Snowie team decided to team up with the community and start a challenge "human vs machine".
The terms of the challenge were simple enough :
- each challenger should play HU, 100+ bb deep, 3 000 hands or more vs Snowie AI
- people with a positive winrate at the end would receive a 1-year PokerSnowie licence
There were 7 challengers to take up the challenge. Most of them 6max / HU regs NL50/NL100 €, playing poker for years.
I was one who participated in this challenge. I didn't start playing right away but only after 1 or 2 day.
After the first night of the challenge, one already started complaining about the unbelievable number of coolers he ran into.
After playing it for some time, I had the exact same feeling, the few times I had a decent hand, he had air and folded to any aggression, but most times when he had something big, I also had a little something (FD on flop, DP or trips on river, etc...).
To me, it seemed unlucky at first, then really weird after more hands, and eventually plain rigged...
Given the fact that there is no way to export our hand history from that program, or to get any information on Snowie stats, etc...
I decided to grab the only stat I could get on my own with a decent sample : the preflop equity.
How do I do that, just check/call every street of every hand and take note of who's winning at showdown.
When I'm in SB, we both have a 100% range, and when I'm in BB I have a 100% range while Snowie has let's say a 90% range given that it's folding a few times.
Therefore I estimate its overall range to be 95% and should then have 50.8% equity against my own range.
After more than 700 hands, its equity is roughly 56% !
I know it's not an extremely huge sample, but given the fact that each hand count (unlike when you check how many AA you were dealt during 10kh dealt, for example), I think it's pretty reasonable to think it's starting to converge a little bit.
The likelyhood to get a result this far off from the theoretical equity over 700 coin flip is extremely low (you can check with any coin flip variance simulator online).
700 hands might seem low, but it's not that bad on this kind of stats.
Of course, when bringing up those results, some people defending Snowie came in the thread screaming some random bull**** about the fact that "of course" it can't be rigged, and started to making it personal to try to flood the reasonable questions we bring up into oblivion.
That was really pathetic of him, everybody noticed.
As if rigging it up wouldn't benefit Snowie's marketing team...
But obviously :
- If their bot is loosing versus pretty much anyone, nobody would pay to have its "GTO" advice regarding their plays.
Would you pay a coach who can't even win $ himself ? Here it's even worse, because if Snowie can't even win on a rake-free poker game, it's 100% worthless...
- On the other hand if they can show that it's winning versus a bunch of players, they will attract customers
Among the 7 people participating in the challenge:
- 1 says he doesn't think it's rigged
- 4 say they experienced a number of bad coolers way beyond anything they have ever seen on any poker room they played on, for years
- 2 didn't say anything
2 of them managed to have a positive winrate (and only 1 is also positive in EV bb/100).
Even the one who won with positive EV bb/100 is among those who claim to see a very biased distribution of setup and was able to be positive by exploiting Snowie on spots where he's weak and being extra careful when hitting anything decent... Very funny, really.
Some other facts about Snowie:
1. They state that it contains "no expert knowledge" inside and has computed their approximation of GTO with neuronal networks.
However, on the beginning of the challenge, Snowie limped a lot of hands from SB (really a lot), which means that it was thinking that it's GTO to limp those range of hands.
Couple of days later, it didn't limp anymore hands, ever (and without any notification of software update - an update came later a bit later though).
A while later the support team of Snowie said limping was a bug and that they fixed it. rotfl...
Yeah right, 2 days ago it was GTO, and now it's no way limping is bad. Sounds like "expert knowledge" included in the bot's strategy to me.
=> they're just lying about that
2. The way they "present" results on their website (for their worldwide challenge) is really poor:
The way Snowie classifies players is bad : the only way to be "good" is to play like him (cause it cannot judge by any other standards if your lines are good or not).
Therefore, if you play like him, or close to it, you will obviously have a very very low winrate.
But if you exploit it, then your playing level will go down (because you deviate from its equilibrium), and therefore your results will be drowned into all those bad players results who try snowie just for fun and shove half their hands just to see what happens.
=> those results are therefore meaningless and just a marketing scam.
3. If they wanted to prove their bot is good and nothing's rigged they could :
- move to open source for the "poker room" part of the software
- participate in world poker botting competition and end up crushing every other bots
- allow exporting hands to HEM/PT to let player at least check some basic stuff
- ...
But they're not doing any of those.
Overall it's more than enough to make up my mind on that software and shelve it on the scam category without any doubt.
And that's the best case scenario, worst case being that by using it you're helping them build a bot they will then put online instead of "sharing/selling" the AI... As it's been mentioned in some other threads here I've seen.
Looking forward to any comments,
nik0