Advances in AI changed poker theory with GTO and Nash theorem. Pluribus will no doubt do same for 6max.
I started to publish my findings on 10k hands from Pluribus.
For instance I noticed Pluribus calls a lot from SB where snowie and others mostly 3bet.
Https://pluribus-poker-ai.com

Get your spam link outta here.

Nash equilibria were discovered 60+ years ago.

These AIs are no where even close to GTO.

They just fixed its SB strategy to limp 50% of the time and open 50% of the time. It says so in that link...

Sure Nash theorem dates back to 51. Using solvers to get closer to gto doesn't and is very recent. And of course AI or piosolver or snowie are not finding perfect gto, just less exploitable than humans, which is why they're so popular. And yes less exploitable doesn't mean most profitable. Still important to study.

I think the way Pluribus plays 6max is very relevant to poker theory. 10k hands have been shared.

I don't think sharing some analysis on these is spam, but if it is, sure let's remove it.

Also nothing was fixed 50/50. Pluribus learned by starting with random distribution and gradually dropped limping from every position except the small blind.

No matter what I'm certain study of Pluribus play is worth it.

This is a direct quote from the article:

"I can’t be 100% sure but it looks very much like the range is 63% and then randomizer 50/50 for limp or bet.
We’ve often thought a mixed limp/bet strategy is ideal but so difficult to implement that we resort to bet / fold only. Now with 50% bet/limp known as probably optimal, we can easily replicate and implement."

This last part is also very wrong. If a strategy randomizes 50/50 between limp and raise it isn't any better or any more balanced than if it just opened 100% of the time.

This forum isn't here for you to shamelessly promote yourself. This thread should be deleted.

What's the difference between this and statmanhal linking to his website?

I don't see any ads or product offers on the linked website. It's just a blog. I don't see anything wrong with this. I'd much rather follow a link than have OP re-post an entire article into the forum to discuss.

I see pretty much all the techniques known for 3-bet; the value hands, the bluff hands from the old and new schools and the hands that can be played either way are played both ways.

The best value hands might be a bit stronger than some might expect but it makes sense and when 3-betting many semi-bluffs and bluffs, it is no problem, although it takes more thinking of how to defend that range vs. 4-bets. And as there is also cold calling (like half the time), the ranges are protected with some stronger cold calls.

As the opens are small, there is less money one folds to a 3-bet, if that makes sense (I read from somewhere it folded more to 3-bets than expected).

The open raises seemed small in these ranges, so some weak hands are opened as well as cold called. If it was a tigher or/and bigger open raise, the ranges for all would be different.

This bot has a 5xbb 3-bet OOP, so if there is something different in the ranges it uses (they could be a bit looser if the size would be smaller), that might be one reason.

I just think the more people analyse and dissect Pluribus 10k hands, the better it will be for our knowledge of poker theory.

The 10k hands are available in the science magazine and Kevin wangg published them in PT4 format on reddit.

I do my bit for my own interest and think it is valuable to share. I do my best not to make mistakes on the data, but my interpretation is of course open to improvements.

Like I say the more people do this, the better, and this is as good a forum there is on poker theory... We didn't have previous AI hands public before with libratus or deepstack, so let's use them.

If course I would love to have 1M hands rather than 10k, but I do learn stuff from this sample no matter what.

Thank you for the share, Pluribus does some very interesting plays. I really like some of the bluffs he does for example or the different type of sizings he uses. this will help me a lot to improve my game. On the other side, even if this is already a crushing poker AI how can it be optimal if he doesn’t deviate from his strategy to exploite the weakness of the other players ?

I will presume that you are either new to poker or new to poker theory/terminology. The word "optimal" has a very specific meaning in poker terms.

There is a sticky in this forum as well as several threads which can help you better understand how that term is used in a poker environment.

hello whosnext thank you for the replay after some
researches I now understand why my sentence was wrongly written. yes new to theory and this forum! Playing cashgame micros right now goal is
to grind up the stakes

Welcome!

I look forward to more of your posts and questions.

Hey... you put a lot of work into this!

Thank you for your posts and the link to your website.

Some posters on this site can be quite rude while you post with class and style. I hope a few bad eggs do not run you off.

Looking forward to reading your future posts and I am sure others are as well.

Thank you, again.

https://www.wsj.com/articles/compute...ly-11562873541

Can anyone access this article?

In the google search under the link the brief description mentions a hand where pluribus bluffs for $2,400 and the only human left remaining is Linus and he makes the call. Maybe someone can dig up and match this HH to figure out who Linus was.

[QUOTE=Pierrecan;55334667]
For instance I noticed Pluribus calls a lot from SB where snowie and others mostly 3bet.

pluribus experiment was a rake-less game

A bit late to this but I looked through the hands and don't find a single hand where Pluribus bets 2400 on river and gets called.

Only instance where Pluribus bets 2400 on river is hand #110016 and Mr. Pink folds his ace-high. (hand below)

-----------

PokerStars Hand #110016: Hold'em No Limit (50/100) - 2019/07/12 06:33:36 ET
Table 'Pluribus Session 110' 6-max (Play Money) Seat #6 is the button
Seat 1: MrPink (10000 in chips)
Seat 2: Pluribus (10000 in chips)
Seat 3: MrWhite (10000 in chips)
Seat 4: MrBlonde (10000 in chips)
Seat 5: MrBlue (10000 in chips)
Seat 6: MrOrange (10000 in chips)
MrPink: posts small blind 50
Pluribus: posts big blind 100
*** HOLE CARDS ***
Dealt to MrPink [Qs Ac]
Dealt to Pluribus [6c Ah]
Dealt to MrWhite [Qc 3c]
Dealt to MrBlonde [2c 7s]
Dealt to MrBlue [9h 2d]
Dealt to MrOrange [Th 4s]
MrWhite: folds
MrBlonde: folds
MrBlue: folds
MrOrange: folds
MrPink: calls 50
Pluribus: checks
*** FLOP *** [3d 4h Jh]
MrPink: bets 100
Pluribus: calls 100
*** TURN *** [3d 4h Jh] [5s]
MrPink: checks
Pluribus: bets 200
MrPink: calls 200
*** RIVER *** [3d 4h Jh] [5s] [Qd]
MrPink: checks
Pluribus: bets 2400
MrPink: folds
Uncalled bet (2400) returned to Pluribus
Pluribus collected 800.0 from pot
*** SUMMARY ***
Total pot 800 | Rake 0
Board [3d 4h Jh 5s Qd]

----------

Anyone ever find out the players' indentities?