thanks. looking forward to it. i tried a few times unsuccessfully myself.

https://en.wikipedia.org/wiki/Eudaemons

Add in the advances in communication and computer technology since the 70s.

Some reasons why 65s is better than other SCs (98s in particular) from one of my favourite posts on 2p2

great stuff

Also 65s is the Empire hand, but I just didn't know if Pluribus was factoring rake-free-card equity in or not.

But yeah, can get on board with explanations of that VPIP tbh. It being part of a less obviously linear range would have to be some of the reason.

This is because if Pluribus was flatting all hands "better than" or equal to 65s at 100% freq, it would surely be exploitable by opening itself up to getting squeezed a lot. If it's a mix, then it's necessarily a 0EV VPIP (indifferent to a 0EV fold), which is probably about as profitable as I could give it credit for being tbh.

Yeah, I actually don't disagree - it's a decently interesting hand. My point is more that we could see these sorts of strategies from a regular HU solver with some reasonable ranges, it's not especially specific to a 6-max AI.

Also the captions from someone who clearly does not know very much about poker just tilts me a bit, not gonna lie.

Hey, I spent the weekend and this whole night writing a converter for pluribus hands to a format that can be imported in poker tracker

Hopefully there are no major bugs left, but I'm sure there are still a couple left.

You can download all the converted hands at the top of the page here. Link to the source code of the converter is also in there. Alright I'm going to bed now since I've been up for a while. Hope someone finds it useful!

Nice one, I know how I'm going to be wasting my afternoon now! Absolute hero.

I also can barely wait to click on an anonymous link (supplied by a guy making his first post ever) which promises it will download things (including "source code") to my laptop.... Great idea!!

lmao, I can recognise a hh file without having to worry about any malicious gremlins - it'll be fine hahaha.

Early conclusion - it's stats are overall more "solid" by the current standards than the supposed "elite pros" it was playing against. I genuinely don't think it's superhuman, but it is undoubtedly a strong AI (as Snowie has been for some time).

thanks for doing this, you have an interesting site Kevin, am checking out the rest of it

These are indeed the correct results? -7bb/100 rake free just very far from solid winning. I understand they did that variance reduction mumbo jumbo and 10k hands across formats is basically a lol sample but let's be realistic here, either it's balance is up and it's won or down and it's lost. Would be nice to see it's EV line and StdDev/100 in PT4.

Thanks a lot for this, really good to have this open source.

Is there a way to tell which aliases are the bots and which are humans? Would also be nice to know which alias is which human , don't know if they revealed this.

Is the 10k hands all that it ever played vs humans so that only a fraction of that was against Linus? Thought it "broke even" vs him over 10k hands (=100k hands with var. reduction), lot less scary if only some were against him.

nvm just saw that Linus didn't even play 5 bots but one bot with 4 other humans. And bot broke even vs Linus and 4 sub top 100 players so don't see the reason to be scared just yet.

Still if someone can divine which alias Linus was that would be great.

Would also be sick to get 10k hands that Ferguson and Elias played against 5 bots.

Beat me to it! I've been working on it pretty solidly yesterday and today. Nice job.

Great job - spent a while looking at some of the hands and the first conclusion is that some of the human play is absolutely terrible, even from my perspective.

Should be able to figure out who Linus is by looking at this nonsense but I'm not going to out him.

I'd just like to confirm that the results for Pluribus provided by mtndew do match exactly with what I've got from my conversions of the hands (showing Pluribus as actually slightly losing overall). So unless we've both somehow made the same mistake in converting the hands Pluribus actually lost in terms of chips over the challenge. I'll attach a photo of the results from my data here (I'm pretty sure they're identical).
https://gyazo.com/37ab64134613d033ac2fbac43367bbc9

Yeah in total I think there were 13 human players in all in the 5 humans plus 1 AI challenge (some of whom played only a thousand or so hands) so it's very possible Linus was beating the bot.

I actually emailed Noam Brown about the hand histories and he said there are no plans to release other hand histories or demos of the bot etc, so I think we're unfortunately unlikely to ever see those other hands.

Personally I'd have liked to a file with millions of hands which Pluribus played against itself

until proven otherwise this bot is Snowie with sensible river play

Can someone please check and report Snowie RFI% for minraise at highest stakes in LJ, HJ, CO. Pluribus has 18%, 24%, 31% respectively which seems too high but I never had a 2x RFI strategy so idk.

RFI frequency are same as preflop solver sims without rake and pretty much what nosebleeds regs raise. Preflop game looks very solid and similar to preflop solvers.

I've seen a couple of turn and river plays that looked weird, and was -EV when I tried to look at them with solver. Of course input=output but I am not convinced it's playing perfectly everywhere.

I think what we are reading is mostly marketing for more research funding.

I take all the claims with more than a few grains of salt.

Half the game of academic research is to continue getting funding. The more fantastic the claims the more money. Kind of like the whole "statistical tie" thing with the claudico challenge.

Do I think they are making progress, YES
BUT I think it is highly exaggerated

Agree with both points. Don't have Snowie but these seemed pretty notably wide to me too. However, I am probably biased having only ever played in formats with substantial rake, so this very easily could be right when there is none.

Also worth noting that I've seen Pluribus use 2x, 2.25x and 2.5x LJ opens in different hh's when clicking through.

Snowie does the same thing.

I think because a hand like 98s can be dominated much more often.
Be it by 9x or AK and K9 on flops like QJT

whereas 56 only gets dominated by JTss on a straight board and A5/A6 on 55x/66x boards.

Actually, just spotted a 3.5x, so it's all over the shop.

PokerStars Hand #83017: Hold'em No Limit (50/100) - 2019/07/11 23:03:37 ET
Table 'Pluribus Session 83' 6-max (Play Money) Seat #6 is the button
Seat 1: Pluribus (10000 in chips)
Seat 2: MrWhite (10000 in chips)
Seat 3: Budd (10000 in chips)
Seat 4: MrOrange (10000 in chips)
Seat 5: Bill (10000 in chips)
Seat 6: MrBlue (10000 in chips)
Pluribus: posts small blind 50
MrWhite: posts big blind 100
*** HOLE CARDS ***
Dealt to Pluribus [Kc As]
Dealt to MrWhite [3h Jc]
Dealt to Budd [5c Ks]
Dealt to MrOrange [4d 2h]
Dealt to Bill [2s 9c]
Dealt to MrBlue [2c 4s]
Budd: calls 100
MrOrange: folds
Bill: folds
MrBlue: folds
Pluribus: raises 1200 to 1300
MrWhite: folds
Budd: folds
Uncalled bet (1200) returned to Pluribus
Pluribus collected 300.0 from pot
*** SUMMARY ***
Total pot 300 | Rake 0
PokerStars Hand #83018: Hold'em No Limit (50/100) - 2019/07/11 23:03:38 ET

Pluribus doesn't **** with limpers