Lol @ the 5-1 offers on bot

If the definition of statistically significant win is something like 95% certainty that actual winrate is >0bb/100, doesn't that mean one needs a ridiculous winrate over that sample size to prove it? I mean almost any reasonably fathomable result (even results that ARE actual >0bb/100 winrates, but too small to fall over the 95% significanec!) over 120k hands is likely within the statistical margin of error, almost rendering the whole thing useless.

No but we have an agreed upon # of hands to play during an agreed upon number of hours so if it takes too long then reaching the desired sample size will not be possible, and that is important to CMU.

It plays instantly and stops to tank and recalculate its strategy at one point during the hand (I believe it's the river again, but just repeating what I was told). I was told that Claudico's avg tank time was 7 seconds and the new agent's avg tank time will be 12 seconds.

bump

it depends on the standard deviation...

Cannot be given ex ante.

the stdev of the match over 120k hands should be fairly easy to estimate, cant be that different from the previous match, huNL isnt that wide of a range

stdev last time was 180 bb/100

For obvious reasons, the "statistical significance" threshold should be determined before the match begins.

A less-than-perfect threshold announced before play commences is infinitely preferred to any threshold announced after the play ends.

^^ from the riverscasino site:

So this is a much looser standard than would be required to publish in an academic journal, no? Last year I think they used the normal 95% confidence standard, which a lot of observers thought was too stringent (at least for the competition aspect of it).

Someone mentioned Chad Powers playing, He's banned from the Rivers.

Is this big bets per 100 (as per PT/HEM), or big blinds per 100? Likewise, for your 9bb/100 win rate in the last competition, was that blinds or bets?

big blinds

yep

Good video; especially, as I remember the Claudico battle as my introduction to "poker ambassador" Doug Polk.

Would have loved to have seen you serve as Coach to the human team for this contest, patrolling between the playing areas, carrying a clip board and wearing a "Bear Bryant" hat.

Hope the humans crush and embarrass the computer so badly that CMU can't reference a statistical tie again and starts blaming Russian hackers.

Big bets? What is this, 2007?

Haha...outed as a limit player.

Anyway, if the humans' true win rate is 9bb/100 (as per last competition) and STD is 180bb/100, they are over 95% likely to be ahead after 120k hands according to pokerdope.

Interestingly that paper seems to suggest that the way to exploit Carnegie Mellon bots is to play preflop and flop call or fold and wait for turn/river for the value bets/bluffs as the bot seems to be bad at estimating opponent ranges on last streets.

I have a few questions.

1. Is 180bb/100 std dev. normal for this stakes?
2. WCG mention that he employ a strategy where he would get to the river less often since the bot take longer to think. Was this strategy share by all 4 players or each 1 play a ddifferent strategy.
3. Did you guys try to be as close to GTO as you can while playing or just ttrying to exploit the bot?
4. If all 4 of you use similar strategy as WCG,
how much of this affect your winrate and std dev?
what do you think the std dev or true win rate should be?

5. Would you guys use similar strategy since this time tthere's more hand and longer tank time at the river by the bot but you get to multi ttable ?

Is this match shown or streamed anywhere this time?

200bb deep 180bb/100 std dev sounds pretty normal.

what does 180bb/100 standard dev. mean exactly?

they are idiots that dont understsand or just dont care. doing these challenges will make unbeatable bots sooner. and thats not what we want. they are so bad for poker and will be remembered as that.

if anyone wants ill take computer at 1:4.15 upto 5k

My 1k to your 4.15k

can escrow to whoever

is it gonna be 200 deep again?