Colman is a HU hyper specialist. He would get wrecked at deepstack HUNL. Isildur is good but not world class.

in case the guy is not trolling, Isildur haven't played hunl in years and it's quite possibly his worst game now, I remember some hs pros on Joeys pod saying that 2-7td is his best game (it was a while ago though)

Colman would do fine in deepstack HUNL. watch the chicagojoey podcast with him. people would challenge him to play hu cash games thinking he was some hyper bum and he would destroy them.

Especially true since his response is so wrong.

https://www.facebook.com/JasonLesPok...59405130966120

I'm glad that, as it seems, I was right this time and HUDs are important in HU high stake.
So again, giving the fact that the bot analyzed the game every night after the play and having info about mirror hands, the bot had quite detailed info about the opponents (same as very detailed HUDs without actual displaying numbers, as bot does not need to see it), while humans did not have either HUDs or mirror hands info.

Either I am not understanding what you are saying, or, dare I say it, you are wrong again. Regarding HUD-type information, the humans had the exact same information as the bot. Each night the humans were given all the information on every hand played that day (hole cards included) for all the players (mirrored hands included, of course).

Were the brains given raw HH data or actual %s and stats on 3b, 4b, $wtsd etc.?

raw HH data

Sorry, I understood very well what you said. But you would probably agree that using processed data (such HUDs) gives significant advantage to a player in human vs human play. In the case we discuss the bot had this info by default and humans did not.
Let's compare it with chess. When Kasparov played with Deep Blue he was allowed to use a computer to check debut info and probably some other stuff. Everybody was agree that Kasparov cannot and should not compete wit Deep Blue in memory capacity, e.g. holding database of debut option. They competed in the actual playing chess.
Similar to our case everybody would agree that even average PC could process and hold information much better than human brain. But this differs from the game itself using this processed information.
Otherwise - yes, it seems obviously now that the bot have huge advantage over the humans. I'm not trying to accuse it of cheating.

I may be wrong, but I believe that Libratus did not utilize any HUD-type information. Every hand was played in a vacuum. It did not "adjust" its play during the day's play.

We are only now learning how it adjusted its play at night, but it appears that that too was not based on "HUD-type" information.

Look, everyone agrees that the challenge could have been set-up differently. And that there were elements of the challenge that were "unfavorable" to the humans (long days, no breaks, no HUDs, long turn tanking on virtually every hand, etc.).

Remember, though, that the challenge was designed by academics to be an "experiment" into the bot's efficacy. It was not designed to give the humans the best chance of beating the bot.

I agree with your statements except the first and most important one
I simply cannot imagine any type of information the bot could use which is not HUD-type. If I'm wrong, give an example.
Once again, I'm not trying to accuse the bot in anything as well as not trying to defend humans. I'm trying to realize what "capacity of winning" exist in poker. Or, simply say, what are the things which allowed the bot to get this significant advantage. If it is, as you say, not HUD-related then I'm intriguing even more.
I came to the thread with confidence that something wrong is here. After reading it and making some simple calculations I admitted that this "capacity" might be quite big.
Now I'm trying to figure out what is the nature of it.

https://www.pokernews.com/strategy/g...ghts-22654.htm

I came up across this article before and now I re-read it. There is no answer on what I'm looking for. More than that this final statement from the article support my previous thoughts that GTO cannot give you a big advantage
I would give this simple explanation to my thoughts:
If I play the worst possible, but mathematically correct way (without intentionally dumping chips), like pushing ai every single hand, then my opponent, playing GTO, will get hundreds BB/100 depending on how deep stack we play. If I'll push 50% of my hands then my opponent, playing GNO, will have significantly less then 50% of initial advantage. And so on. And when I'll be somehow close to GTO, playing perfect GTO will give to my opponent just a miserable advantage.
Come back to our case I assumed that the four humans played against Libratus are good enough so the bot would have some advantage, but not significant.
But I was impressed with 14-28BB/100

I honestly don't know what to say at this point. It is a fact that Libratus beat four top HUNL players for more than 14 bb/100 over 120,000 hands (very statistically significant).

You, and undoubtedly many others, are skeptical when you hear such a result. How can that be? The computer must have been cheating. The computer must have had an unfair advantage. The humans must have been playing poorly. The humans must not have taken the competition seriously. There is no way that a GTO strategy could beat the best humans by more than 14 bb/100. The computer must have been using a max-exploitation strategy to win by more than 14 bb/100. Etc.

Those are all "natural" responses (questions, concerns) when you first hear the result. The problem is that we are virtually certain that nothing "nefarious" occurred during the challenge. And why does anyone think that 14 bb/100 at 200bb depth is impossible to achieve?

Of course, our intuition and previous results are only based upon human vs. human match-ups. We really have (had) no way of knowing how "far" from GTO the best HUNL players played. There are several reasons (some quite obvious) why it is difficult for humans to play HUNL GTO. Just realize how large the memory storage was for Libratus's strategy. And it used a super computer to determine its optimal play on each hand. It should go without saying that these aspects of Libratus far exceed what a human brain can do. (And it is quite likely that Libratus itself is "far" from HUNL GTO.)

We will all likely learn a lot more when CMU releases/publishes detailed information on the challenge and on Libratus.

the whole point of the AI is not to play exploitative but rather GTO. it did not take into account the players stats in making a decision. rather at the end of the day it analyzed the hands played to come closer to a GTO solution.

the rest of your posts are tl;dr talking in circles.

this the whole point of gto is your playing uexploitable you dont exploit therefore you strategy is so balanced you dont adjust to anyone you just play your perfectly balanced strategy

Doug will be putting out a $1k HU package soon, that will include the hands played VS Libratus if You buy it right away it will be include in the price Great Value . . .

2 whosnext
Once again I'm not trying to accuse anybody of any nefarious things. I'm trying to understand the facts according to my understanding of the situation.

2 Rich Checkmaker
I think for the purpose of GTO the bot would not need to play with any players at all. GTO is a pure math and to come to that it would probably be quite enough to play those trillions hands which the bot played itself prior the match. The 120K hands playing during the match would not make much difference for the stage of GTO the bot came up to before the match. The fact that the bot analyzed the hands every night means that it was trying to find more exploitable strategy for these particular players than GTO.
Ok then

It certainly does not mean that.

Your understanding of GTO is not quite complete or correct. But let's skip over that for the moment.

Regarding what Libratus was doing each night, please read the link in post #1131.

Edit: gregorio posted too while I was typing.

playing a trillion hands against itself includes folding AA preflop a bunch and testing the value of that verse raising 3betting etc. the value of each of the trillion hands is tiny compared to the value of hands played against competent opponents employing strategies it has not seen yet. a GTO solution might take quadrillion or quintilions of hands or a magnitude of 1000s or millions (or more!!!) hands than it has already seen. the bot is VERY far from GTO. you are overestimating the value of a trillion hands of poker played by a bot trying to find an equilibrium against itself.

if the computer used its hh vs humans to adjust its play, then that's a effectively a "hud" as we know it, a computer obviously doesn't need a display, why is this a topic of discussion

As far as we know, the computer did not adjust its play during the day based upon previous hands or anything else for that matter.

WTF...Computer played nash. It didnt exploit any players. It doesnt care ****ing HUD, isildur, durr or Phil Ivey!!sScientists are fish poker player. Probably they change some parametres. For example take some more low variance bets

But they will tell it soon. But im 99% sure that it wasnt players game specific adjustments, itwas bot itself