The score's here: https://www.twitch.tv/libratus_extra

day one final score


numbers in () are negative , ai up like 3.5 buyins~? they didn't get to play many hands though. Wonder how they will make up for the hand amount since i'm sure they don't want to spend even more than 20 days grinding


vs claudico they were up a bunch the first day, wonder if tommorow libratus is going to wreck them or if they will make the better adjustments/exploits

Do you even polarize bro?

Is there anywhere we can view the past broadcasts or no?

VODs are turned off on twitch so no for now.

It looks like the bot is play a much more reasonable postflop strategy and given it has supercomputer balancing power it's going to be a hard one for the humans this time. Also saw Jason making some hero folds yesterday which I don't really get, the bot is always going to have bluffs.

I turned into twitch yday for a few minutes. There was a question in chat "Did Libratus train vs Claudico?" -- Sandholm's answer was (loosely quoted): "No it did not train vs Claudico, it trained vs Tartanian8. Tartanian8 is a better bot than Claudico. Tartanian8 won the ACPC in 2016." End of topic.

How full of **** is this guy? There's a scientific paper circulating that shows that in order to win this competition, you have to beat bots which are so bad that they would be better off folding all hands whenever they can. This obviously renders the ACPC a completely useless measure of how good a bot is.

No other references, no info on how Tartatian8 did vs Claudico, no exploitability measures for any bot, and finally no idea given on why the brains played Claudico and not the "better bot" in the first competition (even though they might have stated that somewhere last time and I missed it). #bull****science

Sandholm should be in marketing, not AI research. Credit to him that he realizes that the people need bread and games; and when the bread is really dry, you better make sure there are enough games to distract.

Seems like you don't quite understand how exploitability works. Just because a bot is highly exploitable doesn't mean it's bad. Humans are also extremely exploitable. Imagine you'd have to lay out your entire strategy before playing, by how much do you think someone would be able to exploit you? A metric ton. Humans would do no better than open folding if they were to play their nemesis.

Also the other younger guy stated on stream that the main goal of this bot is to beat humans, low exploitability will come later. Actually the same thing happened in limit holdem. The bots first beat humans while still being highly exploitable, low exploitability came quite a few years after that.

Thanks for stating the obvious. You basically explained (once again) why bot vs human poker competitions are a very noisy measure at best; if a bot loses vs humans over a decent sample one might conclude that the bot is far off from equilibirum, if it wins however, it can mean anything since all that has been achieved is beating an opponent who is likely very far off from equilibrium as well. And how close to equlibirum the bot is is the only thing that matters when you develop algorithms in order to solve large imperfect-information games -- esp as a university that has clearly stated that they're trying to use these techniques in other fields as well (if successful).

Also, "Humans would do no better than open folding if they were to play their nemesis." -- proof?

It's fine if you got sold in their approach, but that doesn't mean it's scientifically sound. Do you really buy the first statement? If that was the case, CMU would have developed an exploitative bot based on the first 80k hands competition in order to beat humans -- did they do that? No, ofc not, they developed a new bot that they believe is better, ie closer to equilibirum. So beating humans and lower exploitability is exactly the same goal, given the approach. The only point is that you don't need to be super close to Nash to beat humans, but obv there is still value in computing lower bounds of exploitability in oder to compare bots, to test whether algorithms converge (esp if they don't have a theoretical guarantee of converging) etc.

Look at the other thread (Deepstack) -- why do you think the authors went through the pain of measuring a lower bound of exploitability even though they got a decent sample vs humans?

I'm usually a fan of your posts; hopefully you can reflect a little more on this. Even if I were wrong on all of the above, I still have every right to call someone out who does things like setting up a competition where it's extremely likely -- ex ante -- that most results will be statistically insignificant, or says things like HUNL is more complicated than Go.

GTO heads up play at NLHE would still be highly exploitable. The optimal counter strategy would still win at a decent rate, probably 5bb+/100.

So even if they do come out with a GTO bot, a human could still conceivably beat it.

80k hands is not enough to develop an exploitative strategy only an exploitable one =P.

Hahahahahahaha

this comes at no surprise however from a guy who once created a thread called "Cepheus the "Unbeatable" bot is beatable"

Isnt it kinda significant that AI up 3.5 bi in mirrored hands even tho just little over 2k hands? Inb4 humans get killed for 10bb/100.

I fantasise about this happening and afterwards WCG, Jungle, Sauce and Ike get offered a good amount of money to rek it and they succeed.

Jason and Dong's hands are mirrored so it is kind of a worry that they are both down.

Humans adapt. Give them a day or two more and I think the humans will have a better grasp on some basics that the bot is doing and what to do to combat it properly.

I mean they are thrown into the ring against a very competent opponent (the bot) which does things in a way they've never encountered. They're almost bound to be thrown off balance at first.

Not saying they won't keep losing to it though.

Jason was asked about this yesterday in twitch chat, about losing both sides of the mirror. He said there were a number of times in the Claudico challenges where humans lost both sides for a day as well. Yesterday swarm and dougiedan both won their mirrored sets.

Lol whut. Please explain. You know the definition of GTO, right?

It always amazes how confident people can be in areas where they have absolutely no ***** clue whats going on

He was levelling...

In that case

A game theory optimal solution to NLHE heads up poker simply means that the optimal counter strategy has the lowest winrate of all optimized counter strategies to all solutions.

Playing against a GTO bot you could see for example on a flop of K82r in a pot with 6 bbs when the bot bets 3bbs he has Kx (x%) etc. Against a known solution you would know exactly what the bots range is. And exactly the frequency of different hands. Like on the K82r flop or whatever you know exactly what percentage likelihood each hand is. You really think that knowing exactly what your opponent is doing in NLHE you wouldn't be able to exploit that???????????? Lol.

Cepheus is beatable. Every GTO solution to poker is beatable. A GTO solution to tic-tac-toe is unbeatable. GTO chess checkers go and all complete information games are unbeatable.

What do you guys think the expected winrate of the optimal counter strategy to a GTO nlhe heads up solution is??? You really think its less than 1bb or something?? NLHE is highly complex.

I take Jungleman over any bot any day of the week.

lol not sure if srs

it amazes me too. you are confident in your superiority but at the same time clueless. its amazing.