Quote:
Originally Posted by Josem
The most obvious proof that your claim is false is that the biggest new product category of the last few years (short-handed, short-stacked, jackpot Sit & Go Tournaments) were solved perfectly by mathematics back in the late 2000s - well more than half a decade before the jackpot Sit & Go Tournaments were popularised by Winamax.
The late stages of those SNGs, where the push/fold strategy is close to optimal, were indeed
approximately solved in the 2000s. However, when the stacks are deeper than 8 bb, push/fold is no longer optimal; limps and, if deeper than ~10 bb, minraises have to be included into the strategy. And the difference between push/fold and advanced strategies is significant in terms of the ROI despite being small by deepstack game standards in terms of bb/100 because the bb is a much larger portion of the BI than in deep games.
Due to the variety of bet sizings available, NLH hasn't been solved by bots - they need to make limiting assumptions (so-called betting abstractions) about possible bet sizings. When only a few sizing options are allowed, a bot can be almost unexploitable, however, when arbitrary sizings are allowed, bots are estimated to be exploitable by hundreds of bb/100 (!), i.e. they'd be better off always folding than playing their leaky strategies. (I'll give a citation later when I retrieve the URL from where I downloaded a PDF.)
Of course, this doesn't imply that bots can't beat the strongest humans. They're already doing it HU. However, even for 3-handed play, with limps, non-all-in raises and postflop action allowed, a bot that can beat human
experts is much harder to build because the game tree is much bigger.
A publicly available solver for 3+ handed play costs €499 (for its core functionality, not the crippled trial version that can only solve turn and river spots) and appears to require a stronger-than-average computer to get a solution for a non-HU preflop spot overnight.
It's not that easy to have a positive chip EV in a 1 fish + 1 reg opponent $100 Spin & Go lineup, leave alone beating the rake. That's because $100 regs know the shortstack postflop play maths - with wide preflop ranges! - very deep. As every tenth of the big blind in EV corresponds to a 0.5%+ difference in the ROI, they can't afford to be as exploitable by humans in terms of bb/100 as 100 bb cash players.
And besides, I was talking not about the early 2010s but about a tendency of
the last 1-2 years when poker sites have finally started realising that shortstack winner-take-all / cash game formats will become vulnerable to bots when hardware gets stronger, and they've started adding various 'gimmicks' into their novel poker formats.
Examples are super-KOs (they're not a new thing, but only last year did Stars begin to promote them hard), Win The Button (that requires future game simulation to estimate the extra value of having the button, and the penalty for being in the blinds, next hand!), Hit & Run satellites at Winamax, Sit & Go Hero at Party, and also, maybe surprisingly to you, Blast at 888 (the reason why it's hard is that when 3-4 players are put all in at once in the end, and 2-4 places are paid, ICM is inaccurate because it doesn't account for the possibility of finishing in 2nd by virtue of being 2nd in chips when the chip leader wins the main pot).
Last edited by coon74; 07-12-2017 at 09:00 AM.