How much did libratus donk bet?

It seems with its strength in balancing mixed strategies, it should be much better with donk betting than humans.

Can't remember the exact number (it may be in this thread somewhere or the Reddit AMA though) but Cheet and Dong commented on this, confirming exactly what you are suggesting.

Quote:

You are severely underestimating how far humans and even Libratus is from GTO.

I might. I’m not good poker strategist or theorist. But you, saying that I underestimate it, don’t tell how you estimate that.
Ad far as I know 14BB/100 is enormous, 29BB/100 is ridiculous. I would be glad if you give me an example of such winrate in a real situation.

https://www.highstakesdb.com/profile..._RedBaron.aspx
his won/hand is $13. at $50/$100 that is 13bb/100, at 25/50 its 26bb/100. and as far as we know he is not a robot.

I'm pretty sure WCG beat Sauce for 20bb/100+ in their hunl challenge.

The old bumhunter HU specialists had 40bb/100 in multi year samples. Obviously not the same as actually playing good players, but still noteworthy. After rake thats probably 60bb.

fyp

OK, yeah, I'm convinced. I was wrong. Seems like the bot is really good and humans are far from perfection.
I still have some doubt because 120K hands is not really big number.
What I would do if I would own the bot is to let it play with another one (the same). It would be noteworthy to see if they will come to the equilibrium and what the variance will be.

haha oops

I think there should be another challange 16 strong(isildusr,durr, Daniel Colman....) hu players vs Libratus. And when Libratus wins again 20bb/100 then we cannot talk anymore "too few hands" "players are tired or tilted"

lol wuut

You really need to read up before commenting so much, libratus have played itself for trillions of hands that how it got its strategy to begin with. the strategy it employs is the one that worked the best vs itself, but as the challenge went on it adjusted to things it thought humans did badly i.e. Stuff itself never do.

120k is a big number when they executed mirror hands and ai ev chops

I had read quite a bit about the topic, but may be was not able to keep all circumstances in my mind all the time.
I know that Libratus played a lot itself adjusting and perfecting its algorithm and strategy. What I meant is that I would like to see the result of that play in its current stage.
As for the number of hands - yes, after I placed the previous comment I also thought that with mirror hands it is significant number.
My apology.

Difficult to believe the adjustment explanations I think they are trying to sell something the AI can't do. Don't believe the AI can invent winning strategies, he can calculate but the strategies must be program by humans.

I'm really sorry if I'm looking very annoying or naive here, but I cannot resist to ask. Does anybody knows the technical details of the match? I could not find it.
In particular, was it physically the same bot playing with all four humans? The simple thing is that if I'm playing mirror hands with two opponents, I'm basically know their hands.
I'm not trying to say that the match organizers cheated this way intentionally, but if it was the same bot then after few thousands hands it could realize that.
Even though there wasn't actual bridge between mirror hands during the play, when the bot analyzed the game during night hours it could realize that. That would probably mean that it had very precise info about players, the same as we would have HUDs based on 100% opponent hands. That would be a huge advantage especially considering that humans did not have HUDs at all.
Once again I admit that it could be very naive assumption and I'm not trying to hurt anybody.
Thanks

No, the bot did not know the players' hole cards due to the mirroring. And, to anticipate your next question, the bot did not know the players' hole cards for any other reason (such as the bot actually being the dealer).

Edit: to be clearer, there were actually four physical bots, not just one. And the bots could not physically communicate with each other.

anywhere i can rewatch this? tyvm in advance 4 da link.

WTF??? 😱

Sorry dude but you have no idea what you're talking about. HUDs are important while playing various opponents so you can remember their differences, high stakes players basically don't need a HUD when they're playing a single opponent.

I cannot argue, never played even close to high stake.
Could you please explain though.
They don't need HUds because they know opponents very well? Still, it would be good to have this knowledge in numbers.
Or, they don't need HUds because opponents are changing their ranges anyways? Then they played basically random game in a certain range.
Other reasons?
Also, from all these conversations I got an impression that the bot won because of its amazing ability to adjust its range against opponents.
That's basically what we need HUDs for.
Am I wrong again?

im almost positive dan polk(long live dan!!!) would disgree. so many different stats that you can use against an opponent and having the exact %s is a lot different than what they feel the number would be.

Why would you not choose world class HUNLH players?

Jungleman would disagree.

He is a big believer in using a HUD in heads up play.

You should consider a less offensive response to someone simply asking some questions.

Colman, isildur are not world class players?